X-Git-Url: https://granicus.if.org/sourcecode?a=blobdiff_plain;f=src%2Fbackend%2Fpostmaster%2Fpgstat.c;h=b41a16de44ce86435068597a40e0fa3537ccd08b;hb=77947c51c08179b8bc12347a7fbcb2c8d7908302;hp=04d8890afd6bc0f4246518003dcd97489c578caa;hpb=dbf53e6345b17be375d35043b6a0e64276f87b75;p=postgresql diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c index 04d8890afd..b41a16de44 100644 --- a/src/backend/postmaster/pgstat.c +++ b/src/backend/postmaster/pgstat.c @@ -11,9 +11,9 @@ * - Add a pgstat config column to pg_database, so this * entire thing can be enabled/disabled on a per db basis. * - * Copyright (c) 2001-2005, PostgreSQL Global Development Group + * Copyright (c) 2001-2007, PostgreSQL Global Development Group * - * $PostgreSQL: pgsql/src/backend/postmaster/pgstat.c,v 1.115 2005/12/31 19:39:10 momjian Exp $ + * $PostgreSQL: pgsql/src/backend/postmaster/pgstat.c,v 1.156 2007/05/27 03:50:39 tgl Exp $ * ---------- */ #include "postgres.h" @@ -28,12 +28,21 @@ #include #include #include +#ifdef HAVE_POLL_H +#include +#endif +#ifdef HAVE_SYS_POLL_H +#include +#endif #include "pgstat.h" #include "access/heapam.h" +#include "access/transam.h" +#include "access/twophase_rmgr.h" #include "access/xact.h" #include "catalog/pg_database.h" +#include "libpq/ip.h" #include "libpq/libpq.h" #include "libpq/pqsignal.h" #include "mb/pg_wchar.h" @@ -46,13 +55,8 @@ #include "storage/ipc.h" #include "storage/pg_shmem.h" #include "storage/pmsignal.h" -#include "storage/procarray.h" -#include "tcop/tcopprot.h" -#include "utils/hsearch.h" #include "utils/memutils.h" #include "utils/ps_status.h" -#include "utils/rel.h" -#include "utils/syscache.h" /* ---------- @@ -69,28 +73,19 @@ #define PGSTAT_STAT_INTERVAL 500 /* How often to write the status file; * in milliseconds. */ -#define PGSTAT_DESTROY_DELAY 10000 /* How long to keep destroyed objects - * known, to give delayed UDP packets - * time to arrive; in milliseconds. */ - -#define PGSTAT_DESTROY_COUNT (PGSTAT_DESTROY_DELAY / PGSTAT_STAT_INTERVAL) - #define PGSTAT_RESTART_INTERVAL 60 /* How often to attempt to restart a * failed statistics collector; in * seconds. */ -/* ---------- - * Amount of space reserved in pgstat_recvbuffer(). - * ---------- - */ -#define PGSTAT_RECVBUFFERSZ ((int) (1024 * sizeof(PgStat_Msg))) +#define PGSTAT_SELECT_TIMEOUT 2 /* How often to check for postmaster + * death; in seconds. */ + /* ---------- * The initial size hints for the hash tables used in the collector. * ---------- */ #define PGSTAT_DB_HASH_SIZE 16 -#define PGSTAT_BE_HASH_SIZE 512 #define PGSTAT_TAB_HASH_SIZE 512 @@ -100,50 +95,96 @@ */ bool pgstat_collect_startcollector = true; bool pgstat_collect_resetonpmstart = false; -bool pgstat_collect_querystring = false; bool pgstat_collect_tuplelevel = false; bool pgstat_collect_blocklevel = false; +bool pgstat_collect_querystring = false; + +/* + * BgWriter global statistics counters (unused in other processes). + * Stored directly in a stats message structure so it can be sent + * without needing to copy things around. We assume this inits to zeroes. + */ +PgStat_MsgBgWriter BgWriterStats; /* ---------- * Local data * ---------- */ NON_EXEC_STATIC int pgStatSock = -1; -NON_EXEC_STATIC int pgStatPipe[2] = {-1, -1}; + static struct sockaddr_storage pgStatAddr; -static pid_t pgStatCollectorPid = 0; static time_t last_pgstat_start_time; -static long pgStatNumMessages = 0; - -static bool pgStatRunningInCollector = FALSE; +static bool pgStatRunningInCollector = false; /* - * Place where backends store per-table info to be sent to the collector. - * We store shared relations separately from non-shared ones, to be able to - * send them in separate messages. + * Structures in which backends store per-table info that's waiting to be + * sent to the collector. + * + * NOTE: once allocated, TabStatusArray structures are never moved or deleted + * for the life of the backend. Also, we zero out the t_id fields of the + * contained PgStat_TableStatus structs whenever they are not actively in use. + * This allows relcache pgstat_info pointers to be treated as long-lived data, + * avoiding repeated searches in pgstat_initstats() when a relation is + * repeatedly opened during a transaction. */ -typedef struct TabStatArray +#define TABSTAT_QUANTUM 100 /* we alloc this many at a time */ + +typedef struct TabStatusArray { - int tsa_alloc; /* num allocated */ - int tsa_used; /* num actually used */ - PgStat_MsgTabstat **tsa_messages; /* the array itself */ -} TabStatArray; + struct TabStatusArray *tsa_next; /* link to next array, if any */ + int tsa_used; /* # entries currently used */ + PgStat_TableStatus tsa_entries[TABSTAT_QUANTUM]; /* per-table data */ +} TabStatusArray; + +static TabStatusArray *pgStatTabList = NULL; -#define TABSTAT_QUANTUM 4 /* we alloc this many at a time */ +/* + * Tuple insertion/deletion counts for an open transaction can't be propagated + * into PgStat_TableStatus counters until we know if it is going to commit + * or abort. Hence, we keep these counts in per-subxact structs that live + * in TopTransactionContext. This data structure is designed on the assumption + * that subxacts won't usually modify very many tables. + */ +typedef struct PgStat_SubXactStatus +{ + int nest_level; /* subtransaction nest level */ + struct PgStat_SubXactStatus *prev; /* higher-level subxact if any */ + PgStat_TableXactStatus *first; /* head of list for this subxact */ +} PgStat_SubXactStatus; -static TabStatArray RegularTabStat = {0, 0, NULL}; -static TabStatArray SharedTabStat = {0, 0, NULL}; +static PgStat_SubXactStatus *pgStatXactStack = NULL; static int pgStatXactCommit = 0; static int pgStatXactRollback = 0; -static TransactionId pgStatDBHashXact = InvalidTransactionId; +/* Record that's written to 2PC state file when pgstat state is persisted */ +typedef struct TwoPhasePgStatRecord +{ + PgStat_Counter tuples_inserted; /* tuples inserted in xact */ + PgStat_Counter tuples_deleted; /* tuples deleted in xact */ + Oid t_id; /* table's OID */ + bool t_shared; /* is it a shared catalog? */ +} TwoPhasePgStatRecord; + +/* + * Info about current "snapshot" of stats file + */ +static MemoryContext pgStatLocalContext = NULL; static HTAB *pgStatDBHash = NULL; -static HTAB *pgStatBeDead = NULL; -static PgStat_StatBeEntry *pgStatBeTable = NULL; -static int pgStatNumBackends = 0; +static PgBackendStatus *localBackendStatusTable = NULL; +static int localNumBackends = 0; + +/* + * Cluster wide statistics, kept in the stats collector. + * Contains statistics that are not collected per database + * or per table. + */ +static PgStat_GlobalStats globalStats; + +static volatile bool need_exit = false; +static volatile bool need_statwrite = false; /* ---------- @@ -151,40 +192,30 @@ static int pgStatNumBackends = 0; * ---------- */ #ifdef EXEC_BACKEND - -typedef enum STATS_PROCESS_TYPE -{ - STAT_PROC_BUFFER, - STAT_PROC_COLLECTOR -} STATS_PROCESS_TYPE; - -static pid_t pgstat_forkexec(STATS_PROCESS_TYPE procType); -static void pgstat_parseArgs(int argc, char *argv[]); +static pid_t pgstat_forkexec(void); #endif -NON_EXEC_STATIC void PgstatBufferMain(int argc, char *argv[]); NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]); -static void pgstat_recvbuffer(void); static void pgstat_exit(SIGNAL_ARGS); -static void pgstat_die(SIGNAL_ARGS); +static void force_statwrite(SIGNAL_ARGS); static void pgstat_beshutdown_hook(int code, Datum arg); static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create); -static int pgstat_add_backend(PgStat_MsgHdr *msg); -static void pgstat_sub_backend(int procpid); -static void pgstat_drop_database(Oid databaseid); static void pgstat_write_statsfile(void); -static void pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, - PgStat_StatBeEntry **betab, - int *numbackends); +static HTAB *pgstat_read_statsfile(Oid onlydb); static void backend_read_statsfile(void); +static void pgstat_read_current_status(void); + +static void pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg); +static HTAB *pgstat_collect_oids(Oid catalogid); + +static PgStat_TableStatus *get_tabstat_entry(Oid rel_id, bool isshared); + +static void pgstat_setup_memcxt(void); static void pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype); static void pgstat_send(void *msg, int len); -static void pgstat_recv_bestart(PgStat_MsgBestart *msg, int len); -static void pgstat_recv_beterm(PgStat_MsgBeterm *msg, int len); -static void pgstat_recv_activity(PgStat_MsgActivity *msg, int len); static void pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len); static void pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len); static void pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len); @@ -192,6 +223,7 @@ static void pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len); static void pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len); static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len); static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len); +static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len); /* ------------------------------------------------------------ @@ -220,14 +252,16 @@ pgstat_init(void) struct timeval tv; char test_byte; int sel_res; + int tries = 0; #define TESTBYTEVAL ((char) 199) /* - * Force start of collector daemon if something to collect + * Force start of collector daemon if something to collect. Note that + * pgstat_collect_querystring is now an independent facility that does not + * require the collector daemon. */ - if (pgstat_collect_querystring || - pgstat_collect_tuplelevel || + if (pgstat_collect_tuplelevel || pgstat_collect_blocklevel) pgstat_collect_startcollector = true; @@ -280,6 +314,10 @@ pgstat_init(void) continue; #endif + if (++tries > 1) + ereport(LOG, + (errmsg("trying another address for the statistics collector"))); + /* * Create the socket. */ @@ -339,8 +377,12 @@ pgstat_init(void) * rules prevent it). */ test_byte = TESTBYTEVAL; + +retry1: if (send(pgStatSock, &test_byte, 1, 0) != 1) { + if (errno == EINTR) + goto retry1; /* if interrupted, just retry */ ereport(LOG, (errcode_for_socket_access(), errmsg("could not send test message on socket for statistics collector: %m"))); @@ -391,8 +433,11 @@ pgstat_init(void) test_byte++; /* just make sure variable is changed */ +retry2: if (recv(pgStatSock, &test_byte, 1, 0) != 1) { + if (errno == EINTR) + goto retry2; /* if interrupted, just retry */ ereport(LOG, (errcode_for_socket_access(), errmsg("could not receive test message on socket for statistics collector: %m"))); @@ -421,9 +466,8 @@ pgstat_init(void) /* * Set the socket to non-blocking IO. This ensures that if the collector - * falls behind (despite the buffering process), statistics messages will - * be discarded; backends won't block waiting to send messages to the - * collector. + * falls behind, statistics messages will be discarded; backends won't + * block waiting to send messages to the collector. */ if (!pg_set_noblock(pgStatSock)) { @@ -450,7 +494,6 @@ startup_failed: /* Adjust GUC variables to suppress useless activity */ pgstat_collect_startcollector = false; - pgstat_collect_querystring = false; pgstat_collect_tuplelevel = false; pgstat_collect_blocklevel = false; } @@ -473,65 +516,23 @@ pgstat_reset_all(void) /* * pgstat_forkexec() - * - * Format up the arglist for, then fork and exec, statistics - * (buffer and collector) processes + * Format up the arglist for, then fork and exec, statistics collector process */ static pid_t -pgstat_forkexec(STATS_PROCESS_TYPE procType) +pgstat_forkexec(void) { char *av[10]; - int ac = 0, - bufc = 0, - i; - char pgstatBuf[2][32]; + int ac = 0; av[ac++] = "postgres"; - - switch (procType) - { - case STAT_PROC_BUFFER: - av[ac++] = "-forkbuf"; - break; - - case STAT_PROC_COLLECTOR: - av[ac++] = "-forkcol"; - break; - - default: - Assert(false); - } - + av[ac++] = "--forkcol"; av[ac++] = NULL; /* filled in by postmaster_forkexec */ - /* postgres_exec_path is not passed by write_backend_variables */ - av[ac++] = postgres_exec_path; - - /* Add to the arg list */ - Assert(bufc <= lengthof(pgstatBuf)); - for (i = 0; i < bufc; i++) - av[ac++] = pgstatBuf[i]; - av[ac] = NULL; Assert(ac < lengthof(av)); return postmaster_forkexec(ac, av); } - - -/* - * pgstat_parseArgs() - - * - * Extract data from the arglist for exec'ed statistics - * (buffer and collector) processes - */ -static void -pgstat_parseArgs(int argc, char *argv[]) -{ - Assert(argc == 4); - - argc = 3; - StrNCpy(postgres_exec_path, argv[argc++], MAXPGPATH); -} #endif /* EXEC_BACKEND */ @@ -590,14 +591,14 @@ pgstat_start(void) * Okay, fork off the collector. */ #ifdef EXEC_BACKEND - switch ((pgStatPid = pgstat_forkexec(STAT_PROC_BUFFER))) + switch ((pgStatPid = pgstat_forkexec())) #else switch ((pgStatPid = fork_process())) #endif { case -1: ereport(LOG, - (errmsg("could not fork statistics buffer: %m"))); + (errmsg("could not fork statistics collector: %m"))); return 0; #ifndef EXEC_BACKEND @@ -606,10 +607,13 @@ pgstat_start(void) /* Close the postmaster's sockets */ ClosePostmasterPorts(false); + /* Lose the postmaster's on-exit routines */ + on_exit_reset(); + /* Drop our connection to postmaster's shared memory, as well */ PGSharedMemoryDetach(); - PgstatBufferMain(0, NULL); + PgstatCollectorMain(0, NULL); break; #endif @@ -621,51 +625,9 @@ pgstat_start(void) return 0; } - -/* ---------- - * pgstat_beterm() - - * - * Called from postmaster to tell collector a backend terminated. - * ---------- - */ -void -pgstat_beterm(int pid) -{ - PgStat_MsgBeterm msg; - - if (pgStatSock < 0) - return; - - /* can't use pgstat_setheader() because it's not called in a backend */ - MemSet(&(msg.m_hdr), 0, sizeof(msg.m_hdr)); - msg.m_hdr.m_type = PGSTAT_MTYPE_BETERM; - msg.m_hdr.m_procpid = pid; - - pgstat_send(&msg, sizeof(msg)); -} - - -/* ---------- - * pgstat_report_autovac() - - * - * Called from autovacuum.c to report startup of an autovacuum process. - * We are called before InitPostgres is done, so can't rely on MyDatabaseId; - * the db OID must be passed in, instead. - * ---------- - */ -void -pgstat_report_autovac(Oid dboid) +void allow_immediate_pgstat_restart(void) { - PgStat_MsgAutovacStart msg; - - if (pgStatSock < 0) - return; - - pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_AUTOVAC_START); - msg.m_databaseid = dboid; - msg.m_start_time = GetCurrentTimestamp(); - - pgstat_send(&msg, sizeof(msg)); + last_pgstat_start_time = 0; } /* ------------------------------------------------------------ @@ -675,196 +637,138 @@ pgstat_report_autovac(Oid dboid) /* ---------- - * pgstat_bestart() - + * pgstat_report_tabstat() - * - * Tell the collector that this new backend is soon ready to process - * queries. Called from InitPostgres. + * Called from tcop/postgres.c to send the so far collected per-table + * access statistics to the collector. Note that this is called only + * when not within a transaction, so it is fair to use transaction stop + * time as an approximation of current time. * ---------- */ void -pgstat_bestart(void) +pgstat_report_tabstat(bool force) { - PgStat_MsgBestart msg; + /* we assume this inits to all zeroes: */ + static const PgStat_TableCounts all_zeroes; + static TimestampTz last_report = 0; + + TimestampTz now; + PgStat_MsgTabstat regular_msg; + PgStat_MsgTabstat shared_msg; + TabStatusArray *tsa; + int i; - if (pgStatSock < 0) + /* Don't expend a clock check if nothing to do */ + if (pgStatTabList == NULL || + pgStatTabList->tsa_used == 0) return; /* - * We may not have a MyProcPort (eg, if this is the autovacuum process). - * For the moment, punt and don't send BESTART --- would be better to work - * out a clean way of handling "unknown clientaddr". + * Don't send a message unless it's been at least PGSTAT_STAT_INTERVAL + * msec since we last sent one, or the caller wants to force stats out. */ - if (MyProcPort) - { - pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_BESTART); - msg.m_databaseid = MyDatabaseId; - msg.m_userid = GetSessionUserId(); - memcpy(&msg.m_clientaddr, &MyProcPort->raddr, sizeof(msg.m_clientaddr)); - pgstat_send(&msg, sizeof(msg)); - } + now = GetCurrentTransactionStopTimestamp(); + if (!force && + !TimestampDifferenceExceeds(last_report, now, PGSTAT_STAT_INTERVAL)) + return; + last_report = now; /* - * Set up a process-exit hook to ensure we flush the last batch of - * statistics to the collector. + * Scan through the TabStatusArray struct(s) to find tables that actually + * have counts, and build messages to send. We have to separate shared + * relations from regular ones because the databaseid field in the + * message header has to depend on that. */ - on_shmem_exit(pgstat_beshutdown_hook, 0); -} + regular_msg.m_databaseid = MyDatabaseId; + shared_msg.m_databaseid = InvalidOid; + regular_msg.m_nentries = 0; + shared_msg.m_nentries = 0; -/* --------- - * pgstat_report_vacuum() - - * - * Tell the collector about the table we just vacuumed. - * --------- - */ -void -pgstat_report_vacuum(Oid tableoid, bool shared, - bool analyze, PgStat_Counter tuples) -{ - PgStat_MsgVacuum msg; - - if (pgStatSock < 0) - return; - - pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_VACUUM); - msg.m_databaseid = shared ? InvalidOid : MyDatabaseId; - msg.m_tableoid = tableoid; - msg.m_analyze = analyze; - msg.m_tuples = tuples; - pgstat_send(&msg, sizeof(msg)); -} + for (tsa = pgStatTabList; tsa != NULL; tsa = tsa->tsa_next) + { + for (i = 0; i < tsa->tsa_used; i++) + { + PgStat_TableStatus *entry = &tsa->tsa_entries[i]; + PgStat_MsgTabstat *this_msg; + PgStat_TableEntry *this_ent; -/* -------- - * pgstat_report_analyze() - - * - * Tell the collector about the table we just analyzed. - * -------- - */ -void -pgstat_report_analyze(Oid tableoid, bool shared, PgStat_Counter livetuples, - PgStat_Counter deadtuples) -{ - PgStat_MsgAnalyze msg; + /* Shouldn't have any pending transaction-dependent counts */ + Assert(entry->trans == NULL); - if (pgStatSock < 0) - return; + /* + * Ignore entries that didn't accumulate any actual counts, + * such as indexes that were opened by the planner but not used. + */ + if (memcmp(&entry->t_counts, &all_zeroes, + sizeof(PgStat_TableCounts)) == 0) + continue; + /* + * OK, insert data into the appropriate message, and send if full. + */ + this_msg = entry->t_shared ? &shared_msg : ®ular_msg; + this_ent = &this_msg->m_entry[this_msg->m_nentries]; + this_ent->t_id = entry->t_id; + memcpy(&this_ent->t_counts, &entry->t_counts, + sizeof(PgStat_TableCounts)); + if (++this_msg->m_nentries >= PGSTAT_NUM_TABENTRIES) + { + pgstat_send_tabstat(this_msg); + this_msg->m_nentries = 0; + } + } + /* zero out TableStatus structs after use */ + MemSet(tsa->tsa_entries, 0, + tsa->tsa_used * sizeof(PgStat_TableStatus)); + tsa->tsa_used = 0; + } - pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ANALYZE); - msg.m_databaseid = shared ? InvalidOid : MyDatabaseId; - msg.m_tableoid = tableoid; - msg.m_live_tuples = livetuples; - msg.m_dead_tuples = deadtuples; - pgstat_send(&msg, sizeof(msg)); + /* + * Send partial messages. If force is true, make sure that any pending + * xact commit/abort gets counted, even if no table stats to send. + */ + if (regular_msg.m_nentries > 0 || + (force && (pgStatXactCommit > 0 || pgStatXactRollback > 0))) + pgstat_send_tabstat(®ular_msg); + if (shared_msg.m_nentries > 0) + pgstat_send_tabstat(&shared_msg); } /* - * Flush any remaining statistics counts out to the collector at process - * exit. Without this, operations triggered during backend exit (such as - * temp table deletions) won't be counted. + * Subroutine for pgstat_report_tabstat: finish and send a tabstat message */ static void -pgstat_beshutdown_hook(int code, Datum arg) -{ - pgstat_report_tabstat(); -} - - -/* ---------- - * pgstat_report_activity() - - * - * Called from tcop/postgres.c to tell the collector what the backend - * is actually doing (usually "" or the start of the query to - * be executed). - * ---------- - */ -void -pgstat_report_activity(const char *cmd_str) +pgstat_send_tabstat(PgStat_MsgTabstat *tsmsg) { - PgStat_MsgActivity msg; + int n; int len; - if (!pgstat_collect_querystring || pgStatSock < 0) - return; - - len = strlen(cmd_str); - len = pg_mbcliplen(cmd_str, len, PGSTAT_ACTIVITY_SIZE - 1); - - memcpy(msg.m_cmd_str, cmd_str, len); - msg.m_cmd_str[len] = '\0'; - len += offsetof(PgStat_MsgActivity, m_cmd_str) + 1; - - pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ACTIVITY); - pgstat_send(&msg, len); -} - - -/* ---------- - * pgstat_report_tabstat() - - * - * Called from tcop/postgres.c to send the so far collected - * per table access statistics to the collector. - * ---------- - */ -void -pgstat_report_tabstat(void) -{ - int i; - - if (pgStatSock < 0 || - !(pgstat_collect_querystring || - pgstat_collect_tuplelevel || - pgstat_collect_blocklevel)) - { - /* Not reporting stats, so just flush whatever we have */ - RegularTabStat.tsa_used = 0; - SharedTabStat.tsa_used = 0; + /* It's unlikely we'd get here with no socket, but maybe not impossible */ + if (pgStatSock < 0) return; - } /* - * For each message buffer used during the last query set the header - * fields and send it out. + * Report accumulated xact commit/rollback whenever we send a normal + * tabstat message */ - for (i = 0; i < RegularTabStat.tsa_used; i++) + if (OidIsValid(tsmsg->m_databaseid)) { - PgStat_MsgTabstat *tsmsg = RegularTabStat.tsa_messages[i]; - int n; - int len; - - n = tsmsg->m_nentries; - len = offsetof(PgStat_MsgTabstat, m_entry[0]) + - n * sizeof(PgStat_TableEntry); - tsmsg->m_xact_commit = pgStatXactCommit; tsmsg->m_xact_rollback = pgStatXactRollback; pgStatXactCommit = 0; pgStatXactRollback = 0; - - pgstat_setheader(&tsmsg->m_hdr, PGSTAT_MTYPE_TABSTAT); - tsmsg->m_databaseid = MyDatabaseId; - pgstat_send(tsmsg, len); } - RegularTabStat.tsa_used = 0; - - /* Ditto, for shared relations */ - for (i = 0; i < SharedTabStat.tsa_used; i++) + else { - PgStat_MsgTabstat *tsmsg = SharedTabStat.tsa_messages[i]; - int n; - int len; - - n = tsmsg->m_nentries; - len = offsetof(PgStat_MsgTabstat, m_entry[0]) + - n * sizeof(PgStat_TableEntry); - - /* We don't report transaction commit/abort here */ tsmsg->m_xact_commit = 0; tsmsg->m_xact_rollback = 0; - - pgstat_setheader(&tsmsg->m_hdr, PGSTAT_MTYPE_TABSTAT); - tsmsg->m_databaseid = InvalidOid; - pgstat_send(tsmsg, len); } - SharedTabStat.tsa_used = 0; + + n = tsmsg->m_nentries; + len = offsetof(PgStat_MsgTabstat, m_entry[0]) + + n * sizeof(PgStat_TableEntry); + + pgstat_setheader(&tsmsg->m_hdr, PGSTAT_MTYPE_TABSTAT); + pgstat_send(tsmsg, len); } @@ -874,26 +778,18 @@ pgstat_report_tabstat(void) * Will tell the collector about objects he can get rid of. * ---------- */ -int +void pgstat_vacuum_tabstat(void) { - Relation dbrel; - HeapScanDesc dbscan; - HeapTuple dbtup; - Oid *dbidlist; - int dbidalloc; - int dbidused; + HTAB *htab; + PgStat_MsgTabpurge msg; HASH_SEQ_STATUS hstat; PgStat_StatDBEntry *dbentry; PgStat_StatTabEntry *tabentry; - HeapTuple reltup; - int nobjects = 0; - PgStat_MsgTabpurge msg; int len; - int i; if (pgStatSock < 0) - return 0; + return; /* * If not done for this transaction, read the statistics collector stats @@ -902,15 +798,41 @@ pgstat_vacuum_tabstat(void) backend_read_statsfile(); /* - * Lookup our own database entry; if not found, nothing to do. + * Read pg_database and make a list of OIDs of all existing databases + */ + htab = pgstat_collect_oids(DatabaseRelationId); + + /* + * Search the database hash table for dead databases and tell the + * collector to drop them. + */ + hash_seq_init(&hstat, pgStatDBHash); + while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL) + { + Oid dbid = dbentry->databaseid; + + CHECK_FOR_INTERRUPTS(); + + if (hash_search(htab, (void *) &dbid, HASH_FIND, NULL) == NULL) + pgstat_drop_database(dbid); + } + + /* Clean up */ + hash_destroy(htab); + + /* + * Lookup our own database entry; if not found, nothing more to do. */ dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash, (void *) &MyDatabaseId, HASH_FIND, NULL); - if (dbentry == NULL) - return -1; - if (dbentry->tables == NULL) - return 0; + if (dbentry == NULL || dbentry->tables == NULL) + return; + + /* + * Similarly to above, make a list of all known relations in this DB. + */ + htab = pgstat_collect_oids(RelationRelationId); /* * Initialize our messages table counter to zero @@ -923,27 +845,20 @@ pgstat_vacuum_tabstat(void) hash_seq_init(&hstat, dbentry->tables); while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&hstat)) != NULL) { - /* - * Check if this relation is still alive by looking up it's pg_class - * tuple in the system catalog cache. - */ - reltup = SearchSysCache(RELOID, - ObjectIdGetDatum(tabentry->tableid), - 0, 0, 0); - if (HeapTupleIsValid(reltup)) - { - ReleaseSysCache(reltup); + Oid tabid = tabentry->tableid; + + CHECK_FOR_INTERRUPTS(); + + if (hash_search(htab, (void *) &tabid, HASH_FIND, NULL) != NULL) continue; - } /* - * Add this table's Oid to the message + * Not there, so add this table's Oid to the message */ - msg.m_tableid[msg.m_nentries++] = tabentry->tableid; - nobjects++; + msg.m_tableid[msg.m_nentries++] = tabid; /* - * If the message is full, send it out and reinitialize to zero + * If the message is full, send it out and reinitialize to empty */ if (msg.m_nentries >= PGSTAT_NUM_TABPURGE) { @@ -971,62 +886,51 @@ pgstat_vacuum_tabstat(void) pgstat_send(&msg, len); } - /* - * Read pg_database and remember the Oid's of all existing databases - */ - dbidalloc = 256; - dbidused = 0; - dbidlist = (Oid *) palloc(sizeof(Oid) * dbidalloc); + /* Clean up */ + hash_destroy(htab); +} - dbrel = heap_open(DatabaseRelationId, AccessShareLock); - dbscan = heap_beginscan(dbrel, SnapshotNow, 0, NULL); - while ((dbtup = heap_getnext(dbscan, ForwardScanDirection)) != NULL) - { - if (dbidused >= dbidalloc) - { - dbidalloc *= 2; - dbidlist = (Oid *) repalloc((char *) dbidlist, - sizeof(Oid) * dbidalloc); - } - dbidlist[dbidused++] = HeapTupleGetOid(dbtup); - } - heap_endscan(dbscan); - heap_close(dbrel, AccessShareLock); - /* - * Search the database hash table for dead databases and tell the - * collector to drop them as well. - */ - hash_seq_init(&hstat, pgStatDBHash); - while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL) +/* ---------- + * pgstat_collect_oids() - + * + * Collect the OIDs of either all databases or all tables, according to + * the parameter, into a temporary hash table. Caller should hash_destroy + * the result when done with it. + * ---------- + */ +static HTAB * +pgstat_collect_oids(Oid catalogid) +{ + HTAB *htab; + HASHCTL hash_ctl; + Relation rel; + HeapScanDesc scan; + HeapTuple tup; + + memset(&hash_ctl, 0, sizeof(hash_ctl)); + hash_ctl.keysize = sizeof(Oid); + hash_ctl.entrysize = sizeof(Oid); + hash_ctl.hash = oid_hash; + htab = hash_create("Temporary table of OIDs", + PGSTAT_TAB_HASH_SIZE, + &hash_ctl, + HASH_ELEM | HASH_FUNCTION); + + rel = heap_open(catalogid, AccessShareLock); + scan = heap_beginscan(rel, SnapshotNow, 0, NULL); + while ((tup = heap_getnext(scan, ForwardScanDirection)) != NULL) { - Oid dbid = dbentry->databaseid; + Oid thisoid = HeapTupleGetOid(tup); - for (i = 0; i < dbidused; i++) - { - if (dbidlist[i] == dbid) - { - dbid = InvalidOid; - break; - } - } + CHECK_FOR_INTERRUPTS(); - if (dbid != InvalidOid) - { - pgstat_drop_database(dbid); - nobjects++; - } + (void) hash_search(htab, (void *) &thisoid, HASH_ENTER, NULL); } + heap_endscan(scan); + heap_close(rel, AccessShareLock); - /* - * Free the dbid list. - */ - pfree(dbidlist); - - /* - * Tell the caller how many removeable objects we found - */ - return nobjects; + return htab; } @@ -1034,12 +938,11 @@ pgstat_vacuum_tabstat(void) * pgstat_drop_database() - * * Tell the collector that we just dropped a database. - * This is the only message that shouldn't get lost in space. Otherwise - * the collector will keep the statistics for the dead DB until his - * stats file got removed while the postmaster is down. + * (If the message gets lost, we will still clean the dead DB eventually + * via future invocations of pgstat_vacuum_tabstat().) * ---------- */ -static void +void pgstat_drop_database(Oid databaseid) { PgStat_MsgDropdb msg; @@ -1053,6 +956,34 @@ pgstat_drop_database(Oid databaseid) } +/* ---------- + * pgstat_drop_relation() - + * + * Tell the collector that we just dropped a relation. + * (If the message gets lost, we will still clean the dead entry eventually + * via future invocations of pgstat_vacuum_tabstat().) + * ---------- + */ +void +pgstat_drop_relation(Oid relid) +{ + PgStat_MsgTabpurge msg; + int len; + + if (pgStatSock < 0) + return; + + msg.m_tableid[0] = relid; + msg.m_nentries = 1; + + len = offsetof(PgStat_MsgTabpurge, m_tableid[0]) +sizeof(Oid); + + pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_TABPURGE); + msg.m_databaseid = MyDatabaseId; + pgstat_send(&msg, len); +} + + /* ---------- * pgstat_reset_counters() - * @@ -1079,262 +1010,640 @@ pgstat_reset_counters(void) /* ---------- - * pgstat_ping() - + * pgstat_report_autovac() - * - * Send some junk data to the collector to increase traffic. + * Called from autovacuum.c to report startup of an autovacuum process. + * We are called before InitPostgres is done, so can't rely on MyDatabaseId; + * the db OID must be passed in, instead. * ---------- */ void -pgstat_ping(void) +pgstat_report_autovac(Oid dboid) { - PgStat_MsgDummy msg; + PgStat_MsgAutovacStart msg; if (pgStatSock < 0) return; - pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DUMMY); + pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_AUTOVAC_START); + msg.m_databaseid = dboid; + msg.m_start_time = GetCurrentTimestamp(); + pgstat_send(&msg, sizeof(msg)); } -/* - * Enlarge a TabStatArray + +/* --------- + * pgstat_report_vacuum() - + * + * Tell the collector about the table we just vacuumed. + * --------- */ -static void -more_tabstat_space(TabStatArray *tsarr) +void +pgstat_report_vacuum(Oid tableoid, bool shared, + bool analyze, PgStat_Counter tuples) { - PgStat_MsgTabstat *newMessages; - PgStat_MsgTabstat **msgArray; - int newAlloc; - int i; + PgStat_MsgVacuum msg; + + if (pgStatSock < 0 || + !pgstat_collect_tuplelevel) + return; + + pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_VACUUM); + msg.m_databaseid = shared ? InvalidOid : MyDatabaseId; + msg.m_tableoid = tableoid; + msg.m_analyze = analyze; + msg.m_autovacuum = IsAutoVacuumWorkerProcess(); /* is this autovacuum? */ + msg.m_vacuumtime = GetCurrentTimestamp(); + msg.m_tuples = tuples; + pgstat_send(&msg, sizeof(msg)); +} + +/* -------- + * pgstat_report_analyze() - + * + * Tell the collector about the table we just analyzed. + * -------- + */ +void +pgstat_report_analyze(Oid tableoid, bool shared, PgStat_Counter livetuples, + PgStat_Counter deadtuples) +{ + PgStat_MsgAnalyze msg; - AssertArg(PointerIsValid(tsarr)); + if (pgStatSock < 0 || + !pgstat_collect_tuplelevel) + return; - newAlloc = tsarr->tsa_alloc + TABSTAT_QUANTUM; + pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_ANALYZE); + msg.m_databaseid = shared ? InvalidOid : MyDatabaseId; + msg.m_tableoid = tableoid; + msg.m_autovacuum = IsAutoVacuumWorkerProcess(); /* is this autovacuum? */ + msg.m_analyzetime = GetCurrentTimestamp(); + msg.m_live_tuples = livetuples; + msg.m_dead_tuples = deadtuples; + pgstat_send(&msg, sizeof(msg)); +} - /* Create (another) quantum of message buffers */ - newMessages = (PgStat_MsgTabstat *) - MemoryContextAllocZero(TopMemoryContext, - sizeof(PgStat_MsgTabstat) * TABSTAT_QUANTUM); - /* Create or enlarge the pointer array */ - if (tsarr->tsa_messages == NULL) - msgArray = (PgStat_MsgTabstat **) - MemoryContextAlloc(TopMemoryContext, - sizeof(PgStat_MsgTabstat *) * newAlloc); - else - msgArray = (PgStat_MsgTabstat **) - repalloc(tsarr->tsa_messages, - sizeof(PgStat_MsgTabstat *) * newAlloc); +/* ---------- + * pgstat_ping() - + * + * Send some junk data to the collector to increase traffic. + * ---------- + */ +void +pgstat_ping(void) +{ + PgStat_MsgDummy msg; - for (i = 0; i < TABSTAT_QUANTUM; i++) - msgArray[tsarr->tsa_alloc + i] = newMessages++; - tsarr->tsa_messages = msgArray; - tsarr->tsa_alloc = newAlloc; + if (pgStatSock < 0) + return; - Assert(tsarr->tsa_used < tsarr->tsa_alloc); + pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_DUMMY); + pgstat_send(&msg, sizeof(msg)); } + /* ---------- * pgstat_initstats() - * - * Called from various places usually dealing with initialization - * of Relation or Scan structures. The data placed into these - * structures from here tell where later to count for buffer reads, - * scans and tuples fetched. + * Initialize a relcache entry to count access statistics. + * Called whenever a relation is opened. + * + * We assume that a relcache entry's pgstat_info field is zeroed by + * relcache.c when the relcache entry is made; thereafter it is long-lived + * data. We can avoid repeated searches of the TabStatus arrays when the + * same relation is touched repeatedly within a transaction. * ---------- */ void -pgstat_initstats(PgStat_Info *stats, Relation rel) +pgstat_initstats(Relation rel) { Oid rel_id = rel->rd_id; - PgStat_TableEntry *useent; - TabStatArray *tsarr; - PgStat_MsgTabstat *tsmsg; - int mb; - int i; + char relkind = rel->rd_rel->relkind; - /* - * Initialize data not to count at all. - */ - stats->tabentry = NULL; + /* We only count stats for things that have storage */ + if (!(relkind == RELKIND_RELATION || + relkind == RELKIND_INDEX || + relkind == RELKIND_TOASTVALUE)) + { + rel->pgstat_info = NULL; + return; + } if (pgStatSock < 0 || !(pgstat_collect_tuplelevel || pgstat_collect_blocklevel)) + { + /* We're not counting at all */ + rel->pgstat_info = NULL; + return; + } + + /* + * If we already set up this relation in the current transaction, + * nothing to do. + */ + if (rel->pgstat_info != NULL && + rel->pgstat_info->t_id == rel_id) return; - tsarr = rel->rd_rel->relisshared ? &SharedTabStat : &RegularTabStat; + /* Else find or make the PgStat_TableStatus entry, and update link */ + rel->pgstat_info = get_tabstat_entry(rel_id, rel->rd_rel->relisshared); +} + +/* + * get_tabstat_entry - find or create a PgStat_TableStatus entry for rel + */ +static PgStat_TableStatus * +get_tabstat_entry(Oid rel_id, bool isshared) +{ + PgStat_TableStatus *entry; + TabStatusArray *tsa; + TabStatusArray *prev_tsa; + int i; /* - * Search the already-used message slots for this relation. + * Search the already-used tabstat slots for this relation. */ - for (mb = 0; mb < tsarr->tsa_used; mb++) + prev_tsa = NULL; + for (tsa = pgStatTabList; tsa != NULL; prev_tsa = tsa, tsa = tsa->tsa_next) { - tsmsg = tsarr->tsa_messages[mb]; - - for (i = tsmsg->m_nentries; --i >= 0;) + for (i = 0; i < tsa->tsa_used; i++) { - if (tsmsg->m_entry[i].t_id == rel_id) - { - stats->tabentry = (void *) &(tsmsg->m_entry[i]); - return; - } + entry = &tsa->tsa_entries[i]; + if (entry->t_id == rel_id) + return entry; } - if (tsmsg->m_nentries >= PGSTAT_NUM_TABENTRIES) - continue; - - /* - * Not found, but found a message buffer with an empty slot instead. - * Fine, let's use this one. - */ - i = tsmsg->m_nentries++; - useent = &tsmsg->m_entry[i]; - MemSet(useent, 0, sizeof(PgStat_TableEntry)); - useent->t_id = rel_id; - stats->tabentry = (void *) useent; - return; + if (tsa->tsa_used < TABSTAT_QUANTUM) + { + /* + * It must not be present, but we found a free slot instead. + * Fine, let's use this one. We assume the entry was already + * zeroed, either at creation or after last use. + */ + entry = &tsa->tsa_entries[tsa->tsa_used++]; + entry->t_id = rel_id; + entry->t_shared = isshared; + return entry; + } } /* - * If we ran out of message buffers, we just allocate more. + * We ran out of tabstat slots, so allocate more. Be sure they're zeroed. */ - if (tsarr->tsa_used >= tsarr->tsa_alloc) - more_tabstat_space(tsarr); + tsa = (TabStatusArray *) MemoryContextAllocZero(TopMemoryContext, + sizeof(TabStatusArray)); + if (prev_tsa) + prev_tsa->tsa_next = tsa; + else + pgStatTabList = tsa; /* - * Use the first entry of the next message buffer. + * Use the first entry of the new TabStatusArray. */ - mb = tsarr->tsa_used++; - tsmsg = tsarr->tsa_messages[mb]; - tsmsg->m_nentries = 1; - useent = &tsmsg->m_entry[0]; - MemSet(useent, 0, sizeof(PgStat_TableEntry)); - useent->t_id = rel_id; - stats->tabentry = (void *) useent; + entry = &tsa->tsa_entries[tsa->tsa_used++]; + entry->t_id = rel_id; + entry->t_shared = isshared; + return entry; } - -/* ---------- - * pgstat_count_xact_commit() - - * - * Called from access/transam/xact.c to count transaction commits. - * ---------- +/* + * get_tabstat_stack_level - add a new (sub)transaction stack entry if needed */ -void -pgstat_count_xact_commit(void) +static PgStat_SubXactStatus * +get_tabstat_stack_level(int nest_level) { - if (!(pgstat_collect_querystring || - pgstat_collect_tuplelevel || - pgstat_collect_blocklevel)) - return; + PgStat_SubXactStatus *xact_state; + + xact_state = pgStatXactStack; + if (xact_state == NULL || xact_state->nest_level != nest_level) + { + xact_state = (PgStat_SubXactStatus *) + MemoryContextAlloc(TopTransactionContext, + sizeof(PgStat_SubXactStatus)); + xact_state->nest_level = nest_level; + xact_state->prev = pgStatXactStack; + xact_state->first = NULL; + pgStatXactStack = xact_state; + } + return xact_state; +} - pgStatXactCommit++; +/* + * add_tabstat_xact_level - add a new (sub)transaction state record + */ +static void +add_tabstat_xact_level(PgStat_TableStatus *pgstat_info, int nest_level) +{ + PgStat_SubXactStatus *xact_state; + PgStat_TableXactStatus *trans; /* - * If there was no relation activity yet, just make one existing message - * buffer used without slots, causing the next report to tell new - * xact-counters. + * If this is the first rel to be modified at the current nest level, + * we first have to push a transaction stack entry. */ - if (RegularTabStat.tsa_alloc == 0) - more_tabstat_space(&RegularTabStat); + xact_state = get_tabstat_stack_level(nest_level); + + /* Now make a per-table stack entry */ + trans = (PgStat_TableXactStatus *) + MemoryContextAllocZero(TopTransactionContext, + sizeof(PgStat_TableXactStatus)); + trans->nest_level = nest_level; + trans->upper = pgstat_info->trans; + trans->parent = pgstat_info; + trans->next = xact_state->first; + xact_state->first = trans; + pgstat_info->trans = trans; +} + +/* + * pgstat_count_heap_insert - count a tuple insertion + */ +void +pgstat_count_heap_insert(Relation rel) +{ + PgStat_TableStatus *pgstat_info = rel->pgstat_info; - if (RegularTabStat.tsa_used == 0) + if (pgstat_collect_tuplelevel && pgstat_info != NULL) { - RegularTabStat.tsa_used++; - RegularTabStat.tsa_messages[0]->m_nentries = 0; + int nest_level = GetCurrentTransactionNestLevel(); + + /* t_tuples_inserted is nontransactional, so just advance it */ + pgstat_info->t_counts.t_tuples_inserted++; + + /* We have to log the transactional effect at the proper level */ + if (pgstat_info->trans == NULL || + pgstat_info->trans->nest_level != nest_level) + add_tabstat_xact_level(pgstat_info, nest_level); + + pgstat_info->trans->tuples_inserted++; } } - -/* ---------- - * pgstat_count_xact_rollback() - - * - * Called from access/transam/xact.c to count transaction rollbacks. - * ---------- +/* + * pgstat_count_heap_update - count a tuple update */ void -pgstat_count_xact_rollback(void) +pgstat_count_heap_update(Relation rel) { - if (!(pgstat_collect_querystring || - pgstat_collect_tuplelevel || - pgstat_collect_blocklevel)) - return; + PgStat_TableStatus *pgstat_info = rel->pgstat_info; - pgStatXactRollback++; + if (pgstat_collect_tuplelevel && pgstat_info != NULL) + { + int nest_level = GetCurrentTransactionNestLevel(); - /* - * If there was no relation activity yet, just make one existing message - * buffer used without slots, causing the next report to tell new - * xact-counters. - */ - if (RegularTabStat.tsa_alloc == 0) - more_tabstat_space(&RegularTabStat); + /* t_tuples_updated is nontransactional, so just advance it */ + pgstat_info->t_counts.t_tuples_updated++; - if (RegularTabStat.tsa_used == 0) + /* We have to log the transactional effect at the proper level */ + if (pgstat_info->trans == NULL || + pgstat_info->trans->nest_level != nest_level) + add_tabstat_xact_level(pgstat_info, nest_level); + + /* An UPDATE both inserts a new tuple and deletes the old */ + pgstat_info->trans->tuples_inserted++; + pgstat_info->trans->tuples_deleted++; + } +} + +/* + * pgstat_count_heap_delete - count a tuple deletion + */ +void +pgstat_count_heap_delete(Relation rel) +{ + PgStat_TableStatus *pgstat_info = rel->pgstat_info; + + if (pgstat_collect_tuplelevel && pgstat_info != NULL) { - RegularTabStat.tsa_used++; - RegularTabStat.tsa_messages[0]->m_nentries = 0; + int nest_level = GetCurrentTransactionNestLevel(); + + /* t_tuples_deleted is nontransactional, so just advance it */ + pgstat_info->t_counts.t_tuples_deleted++; + + /* We have to log the transactional effect at the proper level */ + if (pgstat_info->trans == NULL || + pgstat_info->trans->nest_level != nest_level) + add_tabstat_xact_level(pgstat_info, nest_level); + + pgstat_info->trans->tuples_deleted++; } } /* ---------- - * pgstat_fetch_stat_dbentry() - + * AtEOXact_PgStat * - * Support function for the SQL-callable pgstat* functions. Returns - * the collected statistics for one database or NULL. NULL doesn't mean - * that the database doesn't exist, it is just not yet known by the - * collector, so the caller is better off to report ZERO instead. + * Called from access/transam/xact.c at top-level transaction commit/abort. * ---------- */ -PgStat_StatDBEntry * -pgstat_fetch_stat_dbentry(Oid dbid) +void +AtEOXact_PgStat(bool isCommit) { + PgStat_SubXactStatus *xact_state; + /* - * If not done for this transaction, read the statistics collector stats - * file into some hash tables. + * Count transaction commit or abort. (We use counters, not just bools, + * in case the reporting message isn't sent right away.) */ - backend_read_statsfile(); + if (isCommit) + pgStatXactCommit++; + else + pgStatXactRollback++; /* - * Lookup the requested database; return NULL if not found + * Transfer transactional insert/update counts into the base tabstat + * entries. We don't bother to free any of the transactional state, + * since it's all in TopTransactionContext and will go away anyway. */ - return (PgStat_StatDBEntry *) hash_search(pgStatDBHash, - (void *) &dbid, - HASH_FIND, NULL); -} + xact_state = pgStatXactStack; + if (xact_state != NULL) + { + PgStat_TableXactStatus *trans; + + Assert(xact_state->nest_level == 1); + Assert(xact_state->prev == NULL); + for (trans = xact_state->first; trans != NULL; trans = trans->next) + { + PgStat_TableStatus *tabstat; + + Assert(trans->nest_level == 1); + Assert(trans->upper == NULL); + tabstat = trans->parent; + Assert(tabstat->trans == trans); + if (isCommit) + { + tabstat->t_counts.t_new_live_tuples += trans->tuples_inserted; + tabstat->t_counts.t_new_dead_tuples += trans->tuples_deleted; + } + else + { + /* inserted tuples are dead, deleted tuples are unaffected */ + tabstat->t_counts.t_new_dead_tuples += trans->tuples_inserted; + } + tabstat->trans = NULL; + } + } + pgStatXactStack = NULL; + /* Make sure any stats snapshot is thrown away */ + pgstat_clear_snapshot(); +} /* ---------- - * pgstat_fetch_stat_tabentry() - + * AtEOSubXact_PgStat * - * Support function for the SQL-callable pgstat* functions. Returns - * the collected statistics for one table or NULL. NULL doesn't mean - * that the table doesn't exist, it is just not yet known by the - * collector, so the caller is better off to report ZERO instead. + * Called from access/transam/xact.c at subtransaction commit/abort. * ---------- */ -PgStat_StatTabEntry * -pgstat_fetch_stat_tabentry(Oid relid) +void +AtEOSubXact_PgStat(bool isCommit, int nestDepth) { - Oid dbid; - PgStat_StatDBEntry *dbentry; - PgStat_StatTabEntry *tabentry; + PgStat_SubXactStatus *xact_state; /* - * If not done for this transaction, read the statistics collector stats - * file into some hash tables. + * Transfer transactional insert/update counts into the next higher + * subtransaction state. */ - backend_read_statsfile(); + xact_state = pgStatXactStack; + if (xact_state != NULL && + xact_state->nest_level >= nestDepth) + { + PgStat_TableXactStatus *trans; + PgStat_TableXactStatus *next_trans; - /* - * Lookup our database, then look in its table hash table. - */ - dbid = MyDatabaseId; - dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash, - (void *) &dbid, - HASH_FIND, NULL); + /* delink xact_state from stack immediately to simplify reuse case */ + pgStatXactStack = xact_state->prev; + + for (trans = xact_state->first; trans != NULL; trans = next_trans) + { + PgStat_TableStatus *tabstat; + + next_trans = trans->next; + Assert(trans->nest_level == nestDepth); + tabstat = trans->parent; + Assert(tabstat->trans == trans); + if (isCommit) + { + if (trans->upper && trans->upper->nest_level == nestDepth - 1) + { + trans->upper->tuples_inserted += trans->tuples_inserted; + trans->upper->tuples_deleted += trans->tuples_deleted; + tabstat->trans = trans->upper; + pfree(trans); + } + else + { + /* + * When there isn't an immediate parent state, we can + * just reuse the record instead of going through a + * palloc/pfree pushup (this works since it's all in + * TopTransactionContext anyway). We have to re-link + * it into the parent level, though, and that might mean + * pushing a new entry into the pgStatXactStack. + */ + PgStat_SubXactStatus *upper_xact_state; + + upper_xact_state = get_tabstat_stack_level(nestDepth - 1); + trans->next = upper_xact_state->first; + upper_xact_state->first = trans; + trans->nest_level = nestDepth - 1; + } + } + else + { + /* + * On abort, inserted tuples are dead (and can be bounced out + * to the top-level tabstat), deleted tuples are unaffected + */ + tabstat->t_counts.t_new_dead_tuples += trans->tuples_inserted; + tabstat->trans = trans->upper; + pfree(trans); + } + } + pfree(xact_state); + } +} + + +/* + * AtPrepare_PgStat + * Save the transactional stats state at 2PC transaction prepare. + * + * In this phase we just generate 2PC records for all the pending + * transaction-dependent stats work. + */ +void +AtPrepare_PgStat(void) +{ + PgStat_SubXactStatus *xact_state; + + xact_state = pgStatXactStack; + if (xact_state != NULL) + { + PgStat_TableXactStatus *trans; + + Assert(xact_state->nest_level == 1); + Assert(xact_state->prev == NULL); + for (trans = xact_state->first; trans != NULL; trans = trans->next) + { + PgStat_TableStatus *tabstat; + TwoPhasePgStatRecord record; + + Assert(trans->nest_level == 1); + Assert(trans->upper == NULL); + tabstat = trans->parent; + Assert(tabstat->trans == trans); + + record.tuples_inserted = trans->tuples_inserted; + record.tuples_deleted = trans->tuples_deleted; + record.t_id = tabstat->t_id; + record.t_shared = tabstat->t_shared; + + RegisterTwoPhaseRecord(TWOPHASE_RM_PGSTAT_ID, 0, + &record, sizeof(TwoPhasePgStatRecord)); + } + } +} + +/* + * PostPrepare_PgStat + * Clean up after successful PREPARE. + * + * All we need do here is unlink the transaction stats state from the + * nontransactional state. The nontransactional action counts will be + * reported to the stats collector immediately, while the effects on live + * and dead tuple counts are preserved in the 2PC state file. + * + * Note: AtEOXact_PgStat is not called during PREPARE. + */ +void +PostPrepare_PgStat(void) +{ + PgStat_SubXactStatus *xact_state; + + /* + * We don't bother to free any of the transactional state, + * since it's all in TopTransactionContext and will go away anyway. + */ + xact_state = pgStatXactStack; + if (xact_state != NULL) + { + PgStat_TableXactStatus *trans; + + for (trans = xact_state->first; trans != NULL; trans = trans->next) + { + PgStat_TableStatus *tabstat; + + tabstat = trans->parent; + tabstat->trans = NULL; + } + } + pgStatXactStack = NULL; + + /* Make sure any stats snapshot is thrown away */ + pgstat_clear_snapshot(); +} + +/* + * 2PC processing routine for COMMIT PREPARED case. + * + * Load the saved counts into our local pgstats state. + */ +void +pgstat_twophase_postcommit(TransactionId xid, uint16 info, + void *recdata, uint32 len) +{ + TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata; + PgStat_TableStatus *pgstat_info; + + /* Find or create a tabstat entry for the rel */ + pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared); + + pgstat_info->t_counts.t_new_live_tuples += rec->tuples_inserted; + pgstat_info->t_counts.t_new_dead_tuples += rec->tuples_deleted; +} + +/* + * 2PC processing routine for ROLLBACK PREPARED case. + * + * Load the saved counts into our local pgstats state, but treat them + * as aborted. + */ +void +pgstat_twophase_postabort(TransactionId xid, uint16 info, + void *recdata, uint32 len) +{ + TwoPhasePgStatRecord *rec = (TwoPhasePgStatRecord *) recdata; + PgStat_TableStatus *pgstat_info; + + /* Find or create a tabstat entry for the rel */ + pgstat_info = get_tabstat_entry(rec->t_id, rec->t_shared); + + /* inserted tuples are dead, deleted tuples are no-ops */ + pgstat_info->t_counts.t_new_dead_tuples += rec->tuples_inserted; +} + + +/* ---------- + * pgstat_fetch_stat_dbentry() - + * + * Support function for the SQL-callable pgstat* functions. Returns + * the collected statistics for one database or NULL. NULL doesn't mean + * that the database doesn't exist, it is just not yet known by the + * collector, so the caller is better off to report ZERO instead. + * ---------- + */ +PgStat_StatDBEntry * +pgstat_fetch_stat_dbentry(Oid dbid) +{ + /* + * If not done for this transaction, read the statistics collector stats + * file into some hash tables. + */ + backend_read_statsfile(); + + /* + * Lookup the requested database; return NULL if not found + */ + return (PgStat_StatDBEntry *) hash_search(pgStatDBHash, + (void *) &dbid, + HASH_FIND, NULL); +} + + +/* ---------- + * pgstat_fetch_stat_tabentry() - + * + * Support function for the SQL-callable pgstat* functions. Returns + * the collected statistics for one table or NULL. NULL doesn't mean + * that the table doesn't exist, it is just not yet known by the + * collector, so the caller is better off to report ZERO instead. + * ---------- + */ +PgStat_StatTabEntry * +pgstat_fetch_stat_tabentry(Oid relid) +{ + Oid dbid; + PgStat_StatDBEntry *dbentry; + PgStat_StatTabEntry *tabentry; + + /* + * If not done for this transaction, read the statistics collector stats + * file into some hash tables. + */ + backend_read_statsfile(); + + /* + * Lookup our database, then look in its table hash table. + */ + dbid = MyDatabaseId; + dbentry = (PgStat_StatDBEntry *) hash_search(pgStatDBHash, + (void *) &dbid, + HASH_FIND, NULL); if (dbentry != NULL && dbentry->tables != NULL) { tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables, @@ -1368,40 +1677,365 @@ pgstat_fetch_stat_tabentry(Oid relid) * pgstat_fetch_stat_beentry() - * * Support function for the SQL-callable pgstat* functions. Returns - * the actual activity slot of one active backend. The caller is - * responsible for a check if the actual user is permitted to see - * that info (especially the querystring). + * our local copy of the current-activity entry for one backend. + * + * NB: caller is responsible for a check if the user is permitted to see + * this info (especially the querystring). * ---------- */ -PgStat_StatBeEntry * +PgBackendStatus * pgstat_fetch_stat_beentry(int beid) +{ + pgstat_read_current_status(); + + if (beid < 1 || beid > localNumBackends) + return NULL; + + return &localBackendStatusTable[beid - 1]; +} + + +/* ---------- + * pgstat_fetch_stat_numbackends() - + * + * Support function for the SQL-callable pgstat* functions. Returns + * the maximum current backend id. + * ---------- + */ +int +pgstat_fetch_stat_numbackends(void) +{ + pgstat_read_current_status(); + + return localNumBackends; +} + +/* + * --------- + * pgstat_fetch_global() - + * + * Support function for the SQL-callable pgstat* functions. Returns + * a pointer to the global statistics struct. + * --------- + */ +PgStat_GlobalStats * +pgstat_fetch_global(void) { backend_read_statsfile(); - if (beid < 1 || beid > pgStatNumBackends) - return NULL; + return &globalStats; +} + + +/* ------------------------------------------------------------ + * Functions for management of the shared-memory PgBackendStatus array + * ------------------------------------------------------------ + */ + +static PgBackendStatus *BackendStatusArray = NULL; +static PgBackendStatus *MyBEEntry = NULL; + + +/* + * Report shared-memory space needed by CreateSharedBackendStatus. + */ +Size +BackendStatusShmemSize(void) +{ + Size size; + + size = mul_size(sizeof(PgBackendStatus), MaxBackends); + return size; +} + +/* + * Initialize the shared status array during postmaster startup. + */ +void +CreateSharedBackendStatus(void) +{ + Size size = BackendStatusShmemSize(); + bool found; + + /* Create or attach to the shared array */ + BackendStatusArray = (PgBackendStatus *) + ShmemInitStruct("Backend Status Array", size, &found); + + if (!found) + { + /* + * We're the first - initialize. + */ + MemSet(BackendStatusArray, 0, size); + } +} + + +/* ---------- + * pgstat_bestart() - + * + * Initialize this backend's entry in the PgBackendStatus array, + * and set up an on-proc-exit hook that will clear it again. + * Called from InitPostgres. MyBackendId and MyDatabaseId must be set. + * ---------- + */ +void +pgstat_bestart(void) +{ + volatile PgBackendStatus *beentry; + TimestampTz proc_start_timestamp; + Oid userid; + SockAddr clientaddr; + + Assert(MyBackendId >= 1 && MyBackendId <= MaxBackends); + MyBEEntry = &BackendStatusArray[MyBackendId - 1]; + + /* + * To minimize the time spent modifying the entry, fetch all the needed + * data first. + * + * If we have a MyProcPort, use its session start time (for consistency, + * and to save a kernel call). + */ + if (MyProcPort) + proc_start_timestamp = MyProcPort->SessionStartTime; + else + proc_start_timestamp = GetCurrentTimestamp(); + userid = GetSessionUserId(); + + /* + * We may not have a MyProcPort (eg, if this is the autovacuum process). + * If so, use all-zeroes client address, which is dealt with specially in + * pg_stat_get_backend_client_addr and pg_stat_get_backend_client_port. + */ + if (MyProcPort) + memcpy(&clientaddr, &MyProcPort->raddr, sizeof(clientaddr)); + else + MemSet(&clientaddr, 0, sizeof(clientaddr)); + + /* + * Initialize my status entry, following the protocol of bumping + * st_changecount before and after; and make sure it's even afterwards. We + * use a volatile pointer here to ensure the compiler doesn't try to get + * cute. + */ + beentry = MyBEEntry; + do + { + beentry->st_changecount++; + } while ((beentry->st_changecount & 1) == 0); + + beentry->st_procpid = MyProcPid; + beentry->st_proc_start_timestamp = proc_start_timestamp; + beentry->st_activity_start_timestamp = 0; + beentry->st_txn_start_timestamp = 0; + beentry->st_databaseid = MyDatabaseId; + beentry->st_userid = userid; + beentry->st_clientaddr = clientaddr; + beentry->st_waiting = false; + beentry->st_activity[0] = '\0'; + /* Also make sure the last byte in the string area is always 0 */ + beentry->st_activity[PGBE_ACTIVITY_SIZE - 1] = '\0'; + + beentry->st_changecount++; + Assert((beentry->st_changecount & 1) == 0); + + /* + * Set up a process-exit hook to clean up. + */ + on_shmem_exit(pgstat_beshutdown_hook, 0); +} + +/* + * Shut down a single backend's statistics reporting at process exit. + * + * Flush any remaining statistics counts out to the collector. + * Without this, operations triggered during backend exit (such as + * temp table deletions) won't be counted. + * + * Lastly, clear out our entry in the PgBackendStatus array. + */ +static void +pgstat_beshutdown_hook(int code, Datum arg) +{ + volatile PgBackendStatus *beentry = MyBEEntry; + + pgstat_report_tabstat(true); + + /* + * Clear my status entry, following the protocol of bumping st_changecount + * before and after. We use a volatile pointer here to ensure the + * compiler doesn't try to get cute. + */ + beentry->st_changecount++; + + beentry->st_procpid = 0; /* mark invalid */ + + beentry->st_changecount++; + Assert((beentry->st_changecount & 1) == 0); +} + + +/* ---------- + * pgstat_report_activity() - + * + * Called from tcop/postgres.c to report what the backend is actually doing + * (usually "" or the start of the query to be executed). + * ---------- + */ +void +pgstat_report_activity(const char *cmd_str) +{ + volatile PgBackendStatus *beentry = MyBEEntry; + TimestampTz start_timestamp; + int len; + + if (!pgstat_collect_querystring || !beentry) + return; + + /* + * To minimize the time spent modifying the entry, fetch all the needed + * data first. + */ + start_timestamp = GetCurrentStatementStartTimestamp(); + + len = strlen(cmd_str); + len = pg_mbcliplen(cmd_str, len, PGBE_ACTIVITY_SIZE - 1); + + /* + * Update my status entry, following the protocol of bumping + * st_changecount before and after. We use a volatile pointer here to + * ensure the compiler doesn't try to get cute. + */ + beentry->st_changecount++; + + beentry->st_activity_start_timestamp = start_timestamp; + memcpy((char *) beentry->st_activity, cmd_str, len); + beentry->st_activity[len] = '\0'; + + beentry->st_changecount++; + Assert((beentry->st_changecount & 1) == 0); +} + +/* + * Set the current transaction start timestamp to the specified + * value. If there is no current active transaction, this is signified + * by 0. + */ +void +pgstat_report_txn_timestamp(TimestampTz tstamp) +{ + volatile PgBackendStatus *beentry = MyBEEntry; + + if (!pgstat_collect_querystring || !beentry) + return; + + /* + * Update my status entry, following the protocol of bumping + * st_changecount before and after. We use a volatile pointer + * here to ensure the compiler doesn't try to get cute. + */ + beentry->st_changecount++; + beentry->st_txn_start_timestamp = tstamp; + beentry->st_changecount++; + Assert((beentry->st_changecount & 1) == 0); +} + +/* ---------- + * pgstat_report_waiting() - + * + * Called from lock manager to report beginning or end of a lock wait. + * + * NB: this *must* be able to survive being called before MyBEEntry has been + * initialized. + * ---------- + */ +void +pgstat_report_waiting(bool waiting) +{ + volatile PgBackendStatus *beentry = MyBEEntry; + + if (!pgstat_collect_querystring || !beentry) + return; + + /* + * Since this is a single-byte field in a struct that only this process + * may modify, there seems no need to bother with the st_changecount + * protocol. The update must appear atomic in any case. + */ + beentry->st_waiting = waiting; +} + + +/* ---------- + * pgstat_read_current_status() - + * + * Copy the current contents of the PgBackendStatus array to local memory, + * if not already done in this transaction. + * ---------- + */ +static void +pgstat_read_current_status(void) +{ + volatile PgBackendStatus *beentry; + PgBackendStatus *localtable; + PgBackendStatus *localentry; + int i; + + Assert(!pgStatRunningInCollector); + if (localBackendStatusTable) + return; /* already done */ + + pgstat_setup_memcxt(); + + localtable = (PgBackendStatus *) + MemoryContextAlloc(pgStatLocalContext, + sizeof(PgBackendStatus) * MaxBackends); + localNumBackends = 0; + + beentry = BackendStatusArray; + localentry = localtable; + for (i = 1; i <= MaxBackends; i++) + { + /* + * Follow the protocol of retrying if st_changecount changes while we + * copy the entry, or if it's odd. (The check for odd is needed to + * cover the case where we are able to completely copy the entry while + * the source backend is between increment steps.) We use a volatile + * pointer here to ensure the compiler doesn't try to get cute. + */ + for (;;) + { + int save_changecount = beentry->st_changecount; + + /* + * XXX if PGBE_ACTIVITY_SIZE is really large, it might be best to + * use strcpy not memcpy for copying the activity string? + */ + memcpy(localentry, (char *) beentry, sizeof(PgBackendStatus)); - return &pgStatBeTable[beid - 1]; -} + if (save_changecount == beentry->st_changecount && + (save_changecount & 1) == 0) + break; + /* Make sure we can break out of loop if stuck... */ + CHECK_FOR_INTERRUPTS(); + } -/* ---------- - * pgstat_fetch_stat_numbackends() - - * - * Support function for the SQL-callable pgstat* functions. Returns - * the maximum current backend id. - * ---------- - */ -int -pgstat_fetch_stat_numbackends(void) -{ - backend_read_statsfile(); + beentry++; + /* Only valid entries get included into the local array */ + if (localentry->st_procpid > 0) + { + localentry++; + localNumBackends++; + } + } - return pgStatNumBackends; + /* Set the pointer only after completion of a valid table */ + localBackendStatusTable = localtable; } - /* ------------------------------------------------------------ * Local support functions follow * ------------------------------------------------------------ @@ -1418,8 +2052,6 @@ static void pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype) { hdr->m_type = mtype; - hdr->m_backendid = MyBackendId; - hdr->m_procpid = MyProcPid; } @@ -1432,112 +2064,64 @@ pgstat_setheader(PgStat_MsgHdr *hdr, StatMsgType mtype) static void pgstat_send(void *msg, int len) { + int rc; + if (pgStatSock < 0) return; ((PgStat_MsgHdr *) msg)->m_size = len; + /* We'll retry after EINTR, but ignore all other failures */ + do + { + rc = send(pgStatSock, msg, len, 0); + } while (rc < 0 && errno == EINTR); + #ifdef USE_ASSERT_CHECKING - if (send(pgStatSock, msg, len, 0) < 0) + /* In debug builds, log send failures ... */ + if (rc < 0) elog(LOG, "could not send to statistics collector: %m"); -#else - send(pgStatSock, msg, len, 0); - /* We deliberately ignore any error from send() */ #endif } - /* ---------- - * PgstatBufferMain() - + * pgstat_send_bgwriter() - * - * Start up the statistics buffer process. This is the body of the - * postmaster child process. - * - * The argc/argv parameters are valid only in EXEC_BACKEND case. + * Send bgwriter statistics to the collector * ---------- */ -NON_EXEC_STATIC void -PgstatBufferMain(int argc, char *argv[]) +void +pgstat_send_bgwriter(void) { - IsUnderPostmaster = true; /* we are a postmaster subprocess now */ - - MyProcPid = getpid(); /* reset MyProcPid */ - - /* Lose the postmaster's on-exit routines */ - on_exit_reset(); + /* We assume this initializes to zeroes */ + static const PgStat_MsgBgWriter all_zeroes; /* - * Ignore all signals usually bound to some action in the postmaster, - * except for SIGCHLD and SIGQUIT --- see pgstat_recvbuffer. + * This function can be called even if nothing at all has happened. + * In this case, avoid sending a completely empty message to + * the stats collector. */ - pqsignal(SIGHUP, SIG_IGN); - pqsignal(SIGINT, SIG_IGN); - pqsignal(SIGTERM, SIG_IGN); - pqsignal(SIGQUIT, pgstat_exit); - pqsignal(SIGALRM, SIG_IGN); - pqsignal(SIGPIPE, SIG_IGN); - pqsignal(SIGUSR1, SIG_IGN); - pqsignal(SIGUSR2, SIG_IGN); - pqsignal(SIGCHLD, pgstat_die); - pqsignal(SIGTTIN, SIG_DFL); - pqsignal(SIGTTOU, SIG_DFL); - pqsignal(SIGCONT, SIG_DFL); - pqsignal(SIGWINCH, SIG_DFL); - /* unblock will happen in pgstat_recvbuffer */ - -#ifdef EXEC_BACKEND - pgstat_parseArgs(argc, argv); -#endif + if (memcmp(&BgWriterStats, &all_zeroes, sizeof(PgStat_MsgBgWriter)) == 0) + return; /* - * Start a buffering process to read from the socket, so we have a little - * more time to process incoming messages. - * - * NOTE: the process structure is: postmaster is parent of buffer process - * is parent of collector process. This way, the buffer can detect - * collector failure via SIGCHLD, whereas otherwise it wouldn't notice - * collector failure until it tried to write on the pipe. That would mean - * that after the postmaster started a new collector, we'd have two buffer - * processes competing to read from the UDP socket --- not good. - */ - if (pgpipe(pgStatPipe) < 0) - ereport(ERROR, - (errcode_for_socket_access(), - errmsg("could not create pipe for statistics buffer: %m"))); - - /* child becomes collector process */ -#ifdef EXEC_BACKEND - pgStatCollectorPid = pgstat_forkexec(STAT_PROC_COLLECTOR); -#else - pgStatCollectorPid = fork(); -#endif - switch (pgStatCollectorPid) - { - case -1: - ereport(ERROR, - (errmsg("could not fork statistics collector: %m"))); - -#ifndef EXEC_BACKEND - case 0: - /* child becomes collector process */ - PgstatCollectorMain(0, NULL); - break; -#endif + * Prepare and send the message + */ + pgstat_setheader(&BgWriterStats.m_hdr, PGSTAT_MTYPE_BGWRITER); + pgstat_send(&BgWriterStats, sizeof(BgWriterStats)); - default: - /* parent becomes buffer process */ - closesocket(pgStatPipe[0]); - pgstat_recvbuffer(); - } - exit(0); + /* + * Clear out the statistics buffer, so it can be re-used. + */ + MemSet(&BgWriterStats, 0, sizeof(BgWriterStats)); } /* ---------- * PgstatCollectorMain() - * - * Start up the statistics collector itself. This is the body of the - * postmaster grandchild process. + * Start up the statistics collector process. This is the body of the + * postmaster child process. * * The argc/argv parameters are valid only in EXEC_BACKEND case. * ---------- @@ -1545,34 +2129,44 @@ PgstatBufferMain(int argc, char *argv[]) NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]) { + struct itimerval write_timeout; + bool need_timer = false; + int len; PgStat_Msg msg; + +#ifndef WIN32 +#ifdef HAVE_POLL + struct pollfd input_fd; +#else + struct timeval sel_timeout; fd_set rfds; - int readPipe; - int nready; - int len = 0; - struct timeval timeout; - struct timeval next_statwrite; - bool need_statwrite; - HASHCTL hash_ctl; +#endif +#endif + + IsUnderPostmaster = true; /* we are a postmaster subprocess now */ MyProcPid = getpid(); /* reset MyProcPid */ /* - * Reset signal handling. With the exception of restoring default SIGCHLD - * and SIGQUIT handling, this is a no-op in the non-EXEC_BACKEND case - * because we'll have inherited these settings from the buffer process; - * but it's not a no-op for EXEC_BACKEND. + * If possible, make this process a group leader, so that the postmaster + * can signal any child processes too. (pgstat probably never has + * any child processes, but for consistency we make all postmaster + * child processes do this.) + */ +#ifdef HAVE_SETSID + if (setsid() < 0) + elog(FATAL, "setsid() failed: %m"); +#endif + + /* + * Ignore all signals usually bound to some action in the postmaster, + * except SIGQUIT and SIGALRM. */ pqsignal(SIGHUP, SIG_IGN); pqsignal(SIGINT, SIG_IGN); pqsignal(SIGTERM, SIG_IGN); -#ifndef WIN32 - pqsignal(SIGQUIT, SIG_IGN); -#else - /* kluge to allow buffer process to kill collector; FIXME */ pqsignal(SIGQUIT, pgstat_exit); -#endif - pqsignal(SIGALRM, SIG_IGN); + pqsignal(SIGALRM, force_statwrite); pqsignal(SIGPIPE, SIG_IGN); pqsignal(SIGUSR1, SIG_IGN); pqsignal(SIGUSR2, SIG_IGN); @@ -1583,399 +2177,141 @@ PgstatCollectorMain(int argc, char *argv[]) pqsignal(SIGWINCH, SIG_DFL); PG_SETMASK(&UnBlockSig); -#ifdef EXEC_BACKEND - pgstat_parseArgs(argc, argv); -#endif - - /* Close unwanted files */ - closesocket(pgStatPipe[1]); - closesocket(pgStatSock); - /* * Identify myself via ps */ - init_ps_display("stats collector process", "", ""); - set_ps_display(""); + init_ps_display("stats collector process", "", "", ""); /* * Arrange to write the initial status file right away */ - gettimeofday(&next_statwrite, NULL); - need_statwrite = TRUE; + need_statwrite = true; + + /* Preset the delay between status file writes */ + MemSet(&write_timeout, 0, sizeof(struct itimerval)); + write_timeout.it_value.tv_sec = PGSTAT_STAT_INTERVAL / 1000; + write_timeout.it_value.tv_usec = (PGSTAT_STAT_INTERVAL % 1000) * 1000; /* * Read in an existing statistics stats file or initialize the stats to * zero. */ - pgStatRunningInCollector = TRUE; - pgstat_read_statsfile(&pgStatDBHash, InvalidOid, NULL, NULL); - - /* - * Create the dead backend hashtable - */ - memset(&hash_ctl, 0, sizeof(hash_ctl)); - hash_ctl.keysize = sizeof(int); - hash_ctl.entrysize = sizeof(PgStat_StatBeDead); - hash_ctl.hash = tag_hash; - pgStatBeDead = hash_create("Dead Backends", PGSTAT_BE_HASH_SIZE, - &hash_ctl, HASH_ELEM | HASH_FUNCTION); + pgStatRunningInCollector = true; + pgStatDBHash = pgstat_read_statsfile(InvalidOid); /* - * Create the known backends table + * Setup the descriptor set for select(2). Since only one bit in the set + * ever changes, we need not repeat FD_ZERO each time. */ - pgStatBeTable = (PgStat_StatBeEntry *) - palloc0(sizeof(PgStat_StatBeEntry) * MaxBackends); - - readPipe = pgStatPipe[0]; +#if !defined(HAVE_POLL) && !defined(WIN32) + FD_ZERO(&rfds); +#endif /* - * Process incoming messages and handle all the reporting stuff until - * there are no more messages. + * Loop to process messages until we get SIGQUIT or detect ungraceful + * death of our parent postmaster. + * + * For performance reasons, we don't want to do a PostmasterIsAlive() test + * after every message; instead, do it at statwrite time and if + * select()/poll() is interrupted by timeout. */ for (;;) { + int got_data; + + /* + * Quit if we get SIGQUIT from the postmaster. + */ + if (need_exit) + break; + /* - * If we need to write the status file again (there have been changes - * in the statistics since we wrote it last) calculate the timeout - * until we have to do so. + * If time to write the stats file, do so. Note that the alarm + * interrupt isn't re-enabled immediately, but only after we next + * receive a stats message; so no cycles are wasted when there is + * nothing going on. */ if (need_statwrite) { - struct timeval now; + /* Check for postmaster death; if so we'll write file below */ + if (!PostmasterIsAlive(true)) + break; - gettimeofday(&now, NULL); - /* avoid assuming that tv_sec is signed */ - if (now.tv_sec > next_statwrite.tv_sec || - (now.tv_sec == next_statwrite.tv_sec && - now.tv_usec >= next_statwrite.tv_usec)) - { - timeout.tv_sec = 0; - timeout.tv_usec = 0; - } - else - { - timeout.tv_sec = next_statwrite.tv_sec - now.tv_sec; - timeout.tv_usec = next_statwrite.tv_usec - now.tv_usec; - if (timeout.tv_usec < 0) - { - timeout.tv_sec--; - timeout.tv_usec += 1000000; - } - } + pgstat_write_statsfile(); + need_statwrite = false; + need_timer = true; } /* - * Setup the descriptor set for select(2) + * Wait for a message to arrive; but not for more than + * PGSTAT_SELECT_TIMEOUT seconds. (This determines how quickly we will + * shut down after an ungraceful postmaster termination; so it needn't + * be very fast. However, on some systems SIGQUIT won't interrupt the + * poll/select call, so this also limits speed of response to SIGQUIT, + * which is more important.) + * + * We use poll(2) if available, otherwise select(2). + * Win32 has its own implementation. */ - FD_ZERO(&rfds); - FD_SET(readPipe, &rfds); +#ifndef WIN32 +#ifdef HAVE_POLL + input_fd.fd = pgStatSock; + input_fd.events = POLLIN | POLLERR; + input_fd.revents = 0; - /* - * Now wait for something to do. - */ - nready = select(readPipe + 1, &rfds, NULL, NULL, - (need_statwrite) ? &timeout : NULL); - if (nready < 0) + if (poll(&input_fd, 1, PGSTAT_SELECT_TIMEOUT * 1000) < 0) { if (errno == EINTR) continue; ereport(ERROR, (errcode_for_socket_access(), - errmsg("select() failed in statistics collector: %m"))); - } - - /* - * If there are no descriptors ready, our timeout for writing the - * stats file happened. - */ - if (nready == 0) - { - pgstat_write_statsfile(); - need_statwrite = FALSE; - - continue; - } - - /* - * Check if there is a new statistics message to collect. - */ - if (FD_ISSET(readPipe, &rfds)) - { - /* - * We may need to issue multiple read calls in case the buffer - * process didn't write the message in a single write, which is - * possible since it dumps its buffer bytewise. In any case, we'd - * need two reads since we don't know the message length - * initially. - */ - int nread = 0; - int targetlen = sizeof(PgStat_MsgHdr); /* initial */ - bool pipeEOF = false; - - while (nread < targetlen) - { - len = piperead(readPipe, ((char *) &msg) + nread, - targetlen - nread); - if (len < 0) - { - if (errno == EINTR) - continue; - ereport(ERROR, - (errcode_for_socket_access(), - errmsg("could not read from statistics collector pipe: %m"))); - } - if (len == 0) /* EOF on the pipe! */ - { - pipeEOF = true; - break; - } - nread += len; - if (nread == sizeof(PgStat_MsgHdr)) - { - /* we have the header, compute actual msg length */ - targetlen = msg.msg_hdr.m_size; - if (targetlen < (int) sizeof(PgStat_MsgHdr) || - targetlen > (int) sizeof(msg)) - { - /* - * Bogus message length implies that we got out of - * sync with the buffer process somehow. Abort so that - * we can restart both processes. - */ - ereport(ERROR, - (errmsg("invalid statistics message length"))); - } - } - } - - /* - * EOF on the pipe implies that the buffer process exited. Fall - * out of outer loop. - */ - if (pipeEOF) - break; - - /* - * Distribute the message to the specific function handling it. - */ - switch (msg.msg_hdr.m_type) - { - case PGSTAT_MTYPE_DUMMY: - break; - - case PGSTAT_MTYPE_BESTART: - pgstat_recv_bestart((PgStat_MsgBestart *) &msg, nread); - break; - - case PGSTAT_MTYPE_BETERM: - pgstat_recv_beterm((PgStat_MsgBeterm *) &msg, nread); - break; - - case PGSTAT_MTYPE_TABSTAT: - pgstat_recv_tabstat((PgStat_MsgTabstat *) &msg, nread); - break; - - case PGSTAT_MTYPE_TABPURGE: - pgstat_recv_tabpurge((PgStat_MsgTabpurge *) &msg, nread); - break; - - case PGSTAT_MTYPE_ACTIVITY: - pgstat_recv_activity((PgStat_MsgActivity *) &msg, nread); - break; - - case PGSTAT_MTYPE_DROPDB: - pgstat_recv_dropdb((PgStat_MsgDropdb *) &msg, nread); - break; - - case PGSTAT_MTYPE_RESETCOUNTER: - pgstat_recv_resetcounter((PgStat_MsgResetcounter *) &msg, - nread); - break; - - case PGSTAT_MTYPE_AUTOVAC_START: - pgstat_recv_autovac((PgStat_MsgAutovacStart *) &msg, nread); - break; - - case PGSTAT_MTYPE_VACUUM: - pgstat_recv_vacuum((PgStat_MsgVacuum *) &msg, nread); - break; - - case PGSTAT_MTYPE_ANALYZE: - pgstat_recv_analyze((PgStat_MsgAnalyze *) &msg, nread); - break; - - default: - break; - } - - /* - * Globally count messages. - */ - pgStatNumMessages++; - - /* - * If this is the first message after we wrote the stats file the - * last time, setup the timeout that it'd be written. - */ - if (!need_statwrite) - { - gettimeofday(&next_statwrite, NULL); - next_statwrite.tv_usec += ((PGSTAT_STAT_INTERVAL) * 1000); - next_statwrite.tv_sec += (next_statwrite.tv_usec / 1000000); - next_statwrite.tv_usec %= 1000000; - need_statwrite = TRUE; - } + errmsg("poll() failed in statistics collector: %m"))); } - /* - * Note that we do NOT check for postmaster exit inside the loop; only - * EOF on the buffer pipe causes us to fall out. This ensures we - * don't exit prematurely if there are still a few messages in the - * buffer or pipe at postmaster shutdown. - */ - } - - /* - * Okay, we saw EOF on the buffer pipe, so there are no more messages to - * process. If the buffer process quit because of postmaster shutdown, we - * want to save the final stats to reuse at next startup. But if the - * buffer process failed, it seems best not to (there may even now be a - * new collector firing up, and we don't want it to read a - * partially-rewritten stats file). - */ - if (!PostmasterIsAlive(false)) - pgstat_write_statsfile(); -} - - -/* ---------- - * pgstat_recvbuffer() - - * - * This is the body of the separate buffering process. Its only - * purpose is to receive messages from the UDP socket as fast as - * possible and forward them over a pipe into the collector itself. - * If the collector is slow to absorb messages, they are buffered here. - * ---------- - */ -static void -pgstat_recvbuffer(void) -{ - fd_set rfds; - fd_set wfds; - struct timeval timeout; - int writePipe = pgStatPipe[1]; - int maxfd; - int nready; - int len; - int xfr; - int frm; - PgStat_Msg input_buffer; - char *msgbuffer; - int msg_send = 0; /* next send index in buffer */ - int msg_recv = 0; /* next receive index */ - int msg_have = 0; /* number of bytes stored */ - bool overflow = false; - - /* - * Identify myself via ps - */ - init_ps_display("stats buffer process", "", ""); - set_ps_display(""); - - /* - * We want to die if our child collector process does. There are two ways - * we might notice that it has died: receive SIGCHLD, or get a write - * failure on the pipe leading to the child. We can set SIGPIPE to kill - * us here. Our SIGCHLD handler was already set up before we forked (must - * do it that way, else it's a race condition). - */ - pqsignal(SIGPIPE, SIG_DFL); - PG_SETMASK(&UnBlockSig); - - /* - * Set the write pipe to nonblock mode, so that we cannot block when the - * collector falls behind. - */ - if (!pg_set_noblock(writePipe)) - ereport(ERROR, - (errcode_for_socket_access(), - errmsg("could not set statistics collector pipe to nonblocking mode: %m"))); - - /* - * Allocate the message buffer - */ - msgbuffer = (char *) palloc(PGSTAT_RECVBUFFERSZ); - - /* - * Loop forever - */ - for (;;) - { - FD_ZERO(&rfds); - FD_ZERO(&wfds); - maxfd = -1; - - /* - * As long as we have buffer space we add the socket to the read - * descriptor set. - */ - if (msg_have <= (int) (PGSTAT_RECVBUFFERSZ - sizeof(PgStat_Msg))) - { - FD_SET(pgStatSock, &rfds); - maxfd = pgStatSock; - overflow = false; - } - else - { - if (!overflow) - { - ereport(LOG, - (errmsg("statistics buffer is full"))); - overflow = true; - } - } + got_data = (input_fd.revents != 0); +#else /* !HAVE_POLL */ - /* - * If we have messages to write out, we add the pipe to the write - * descriptor set. - */ - if (msg_have > 0) - { - FD_SET(writePipe, &wfds); - if (writePipe > maxfd) - maxfd = writePipe; - } + FD_SET(pgStatSock, &rfds); /* - * Wait for some work to do; but not for more than 10 seconds. (This - * determines how quickly we will shut down after an ungraceful - * postmaster termination; so it needn't be very fast.) + * timeout struct is modified by select() on some operating systems, + * so re-fill it each time. */ - timeout.tv_sec = 10; - timeout.tv_usec = 0; + sel_timeout.tv_sec = PGSTAT_SELECT_TIMEOUT; + sel_timeout.tv_usec = 0; - nready = select(maxfd + 1, &rfds, &wfds, NULL, &timeout); - if (nready < 0) + if (select(pgStatSock + 1, &rfds, NULL, NULL, &sel_timeout) < 0) { if (errno == EINTR) continue; ereport(ERROR, (errcode_for_socket_access(), - errmsg("select() failed in statistics buffer: %m"))); + errmsg("select() failed in statistics collector: %m"))); } + got_data = FD_ISSET(pgStatSock, &rfds); +#endif /* HAVE_POLL */ +#else /* WIN32 */ + got_data = pgwin32_waitforsinglesocket(pgStatSock, FD_READ, + PGSTAT_SELECT_TIMEOUT*1000); +#endif + /* * If there is a message on the socket, read it and check for * validity. */ - if (FD_ISSET(pgStatSock, &rfds)) + if (got_data) { - len = recv(pgStatSock, (char *) &input_buffer, + len = recv(pgStatSock, (char *) &msg, sizeof(PgStat_Msg), 0); if (len < 0) + { + if (errno == EINTR) + continue; ereport(ERROR, (errcode_for_socket_access(), errmsg("could not read statistics message: %m"))); + } /* * We ignore messages that are smaller than our common header @@ -1986,193 +2322,102 @@ pgstat_recvbuffer(void) /* * The received length must match the length in the header */ - if (input_buffer.msg_hdr.m_size != len) + if (msg.msg_hdr.m_size != len) continue; /* - * O.K. - we accept this message. Copy it to the circular - * msgbuffer. + * O.K. - we accept this message. Process it. */ - frm = 0; - while (len > 0) - { - xfr = PGSTAT_RECVBUFFERSZ - msg_recv; - if (xfr > len) - xfr = len; - Assert(xfr > 0); - memcpy(msgbuffer + msg_recv, - ((char *) &input_buffer) + frm, - xfr); - msg_recv += xfr; - if (msg_recv == PGSTAT_RECVBUFFERSZ) - msg_recv = 0; - msg_have += xfr; - frm += xfr; - len -= xfr; - } - } - - /* - * If the collector is ready to receive, write some data into his - * pipe. We may or may not be able to write all that we have. - * - * NOTE: if what we have is less than PIPE_BUF bytes but more than the - * space available in the pipe buffer, most kernels will refuse to - * write any of it, and will return EAGAIN. This means we will - * busy-loop until the situation changes (either because the collector - * caught up, or because more data arrives so that we have more than - * PIPE_BUF bytes buffered). This is not good, but is there any way - * around it? We have no way to tell when the collector has caught - * up... - */ - if (FD_ISSET(writePipe, &wfds)) - { - xfr = PGSTAT_RECVBUFFERSZ - msg_send; - if (xfr > msg_have) - xfr = msg_have; - Assert(xfr > 0); - len = pipewrite(writePipe, msgbuffer + msg_send, xfr); - if (len < 0) + switch (msg.msg_hdr.m_type) { - if (errno == EINTR || errno == EAGAIN) - continue; /* not enough space in pipe */ - ereport(ERROR, - (errcode_for_socket_access(), - errmsg("could not write to statistics collector pipe: %m"))); - } - /* NB: len < xfr is okay */ - msg_send += len; - if (msg_send == PGSTAT_RECVBUFFERSZ) - msg_send = 0; - msg_have -= len; - } - - /* - * Make sure we forwarded all messages before we check for postmaster - * termination. - */ - if (msg_have != 0 || FD_ISSET(pgStatSock, &rfds)) - continue; + case PGSTAT_MTYPE_DUMMY: + break; - /* - * If the postmaster has terminated, we die too. (This is no longer - * the normal exit path, however.) - */ - if (!PostmasterIsAlive(true)) - exit(0); - } -} + case PGSTAT_MTYPE_TABSTAT: + pgstat_recv_tabstat((PgStat_MsgTabstat *) &msg, len); + break; -/* SIGQUIT signal handler for buffer process */ -static void -pgstat_exit(SIGNAL_ARGS) -{ - /* - * For now, we just nail the doors shut and get out of town. It might be - * cleaner to allow any pending messages to be sent, but that creates a - * tradeoff against speed of exit. - */ + case PGSTAT_MTYPE_TABPURGE: + pgstat_recv_tabpurge((PgStat_MsgTabpurge *) &msg, len); + break; - /* - * If running in bufferer, kill our collector as well. On some broken - * win32 systems, it does not shut down automatically because of issues - * with socket inheritance. XXX so why not fix the socket inheritance... - */ -#ifdef WIN32 - if (pgStatCollectorPid > 0) - kill(pgStatCollectorPid, SIGQUIT); -#endif - exit(0); -} + case PGSTAT_MTYPE_DROPDB: + pgstat_recv_dropdb((PgStat_MsgDropdb *) &msg, len); + break; -/* SIGCHLD signal handler for buffer process */ -static void -pgstat_die(SIGNAL_ARGS) -{ - exit(1); -} + case PGSTAT_MTYPE_RESETCOUNTER: + pgstat_recv_resetcounter((PgStat_MsgResetcounter *) &msg, + len); + break; + case PGSTAT_MTYPE_AUTOVAC_START: + pgstat_recv_autovac((PgStat_MsgAutovacStart *) &msg, len); + break; -/* ---------- - * pgstat_add_backend() - - * - * Support function to keep our backend list up to date. - * ---------- - */ -static int -pgstat_add_backend(PgStat_MsgHdr *msg) -{ - PgStat_StatBeEntry *beentry; - PgStat_StatBeDead *deadbe; + case PGSTAT_MTYPE_VACUUM: + pgstat_recv_vacuum((PgStat_MsgVacuum *) &msg, len); + break; - /* - * Check that the backend ID is valid - */ - if (msg->m_backendid < 1 || msg->m_backendid > MaxBackends) - { - ereport(LOG, - (errmsg("invalid server process ID %d", msg->m_backendid))); - return -1; - } + case PGSTAT_MTYPE_ANALYZE: + pgstat_recv_analyze((PgStat_MsgAnalyze *) &msg, len); + break; - /* - * Get the slot for this backendid. - */ - beentry = &pgStatBeTable[msg->m_backendid - 1]; + case PGSTAT_MTYPE_BGWRITER: + pgstat_recv_bgwriter((PgStat_MsgBgWriter *) &msg, len); + break; - /* - * If the slot contains the PID of this backend, everything is fine and we - * have nothing to do. Note that all the slots are zero'd out when the - * collector is started. We assume that a slot is "empty" iff procpid == - * 0. - */ - if (beentry->procpid > 0 && beentry->procpid == msg->m_procpid) - return 0; + default: + break; + } - /* - * Lookup if this backend is known to be dead. This can be caused due to - * messages arriving in the wrong order - e.g. postmaster's BETERM message - * might have arrived before we received all the backends stats messages, - * or even a new backend with the same backendid was faster in sending his - * BESTART. - * - * If the backend is known to be dead, we ignore this add. - */ - deadbe = (PgStat_StatBeDead *) hash_search(pgStatBeDead, - (void *) &(msg->m_procpid), - HASH_FIND, NULL); - if (deadbe) - return 1; + /* + * If this is the first message after we wrote the stats file the + * last time, enable the alarm interrupt to make it be written + * again later. + */ + if (need_timer) + { + if (setitimer(ITIMER_REAL, &write_timeout, NULL)) + ereport(ERROR, + (errmsg("could not set statistics collector timer: %m"))); + need_timer = false; + } + } + else + { + /* + * We can only get here if the select/poll timeout elapsed. Check + * for postmaster death. + */ + if (!PostmasterIsAlive(true)) + break; + } + } /* end of message-processing loop */ /* - * Backend isn't known to be dead. If it's slot is currently used, we have - * to kick out the old backend. + * Save the final stats to reuse at next startup. */ - if (beentry->procpid > 0) - pgstat_sub_backend(beentry->procpid); + pgstat_write_statsfile(); - /* Must be able to distinguish between empty and non-empty slots */ - Assert(msg->m_procpid > 0); + exit(0); +} - /* Put this new backend into the slot */ - beentry->procpid = msg->m_procpid; - beentry->start_timestamp = GetCurrentTimestamp(); - beentry->activity_start_timestamp = 0; - beentry->activity[0] = '\0'; - /* - * We can't initialize the rest of the data in this slot until we see the - * BESTART message. Therefore, we set the database and user to sentinel - * values, to indicate "undefined". There is no easy way to do this for - * the client address, so make sure to check that the database or user are - * defined before accessing the client address. - */ - beentry->userid = InvalidOid; - beentry->databaseid = InvalidOid; +/* SIGQUIT signal handler for collector process */ +static void +pgstat_exit(SIGNAL_ARGS) +{ + need_exit = true; +} - return 0; +/* SIGALRM signal handler for collector process */ +static void +force_statwrite(SIGNAL_ARGS) +{ + need_statwrite = true; } + /* * Lookup the hash table entry for the specified database. If no hash * table entry exists, initialize it, if the create parameter is true. @@ -2203,7 +2448,11 @@ pgstat_get_db_entry(Oid databaseid, bool create) result->n_xact_rollback = 0; result->n_blocks_fetched = 0; result->n_blocks_hit = 0; - result->destroy = 0; + result->n_tuples_returned = 0; + result->n_tuples_fetched = 0; + result->n_tuples_inserted = 0; + result->n_tuples_updated = 0; + result->n_tuples_deleted = 0; result->last_autovac_time = 0; memset(&hash_ctl, 0, sizeof(hash_ctl)); @@ -2219,61 +2468,6 @@ pgstat_get_db_entry(Oid databaseid, bool create) return result; } -/* ---------- - * pgstat_sub_backend() - - * - * Remove a backend from the actual backends list. - * ---------- - */ -static void -pgstat_sub_backend(int procpid) -{ - int i; - PgStat_StatBeDead *deadbe; - bool found; - - /* - * Search in the known-backends table for the slot containing this PID. - */ - for (i = 0; i < MaxBackends; i++) - { - if (pgStatBeTable[i].procpid == procpid) - { - /* - * That's him. Add an entry to the known to be dead backends. Due - * to possible misorder in the arrival of UDP packets it's - * possible that even if we know the backend is dead, there could - * still be messages queued that arrive later. Those messages must - * not cause our number of backends statistics to get screwed up, - * so we remember for a couple of seconds that this PID is dead - * and ignore them (only the counting of backends, not the table - * access stats they sent). - */ - deadbe = (PgStat_StatBeDead *) hash_search(pgStatBeDead, - (void *) &procpid, - HASH_ENTER, - &found); - - if (!found) - { - deadbe->backendid = i + 1; - deadbe->destroy = PGSTAT_DESTROY_COUNT; - } - - /* - * Declare the backend slot empty. - */ - pgStatBeTable[i].procpid = 0; - return; - } - } - - /* - * No big problem if not found. This can happen if UDP messages arrive out - * of order here. - */ -} - /* ---------- * pgstat_write_statsfile() - @@ -2288,9 +2482,7 @@ pgstat_write_statsfile(void) HASH_SEQ_STATUS tstat; PgStat_StatDBEntry *dbentry; PgStat_StatTabEntry *tabentry; - PgStat_StatBeDead *deadbe; FILE *fpout; - int i; int32 format_id; /* @@ -2312,6 +2504,11 @@ pgstat_write_statsfile(void) format_id = PGSTAT_FILE_FORMAT_ID; fwrite(&format_id, sizeof(format_id), 1, fpout); + /* + * Write global stats struct + */ + fwrite(&globalStats, sizeof(globalStats), 1, fpout); + /* * Walk through the database table. */ @@ -2319,68 +2516,19 @@ pgstat_write_statsfile(void) while ((dbentry = (PgStat_StatDBEntry *) hash_seq_search(&hstat)) != NULL) { /* - * If this database is marked destroyed, count down and do so if it - * reaches 0. - */ - if (dbentry->destroy > 0) - { - if (--(dbentry->destroy) == 0) - { - if (dbentry->tables != NULL) - hash_destroy(dbentry->tables); - - if (hash_search(pgStatDBHash, - (void *) &(dbentry->databaseid), - HASH_REMOVE, NULL) == NULL) - ereport(ERROR, - (errmsg("database hash table corrupted " - "during cleanup --- abort"))); - } - - /* - * Don't include statistics for it. - */ - continue; - } - - /* - * Write out the DB line including the number of live backends. + * Write out the DB entry including the number of live backends. We + * don't write the tables pointer since it's of no use to any other + * process. */ fputc('D', fpout); - fwrite(dbentry, sizeof(PgStat_StatDBEntry), 1, fpout); + fwrite(dbentry, offsetof(PgStat_StatDBEntry, tables), 1, fpout); /* - * Walk through the databases access stats per table. + * Walk through the database's access stats per table. */ hash_seq_init(&tstat, dbentry->tables); while ((tabentry = (PgStat_StatTabEntry *) hash_seq_search(&tstat)) != NULL) { - /* - * If table entry marked for destruction, same as above for the - * database entry. - */ - if (tabentry->destroy > 0) - { - if (--(tabentry->destroy) == 0) - { - if (hash_search(dbentry->tables, - (void *) &(tabentry->tableid), - HASH_REMOVE, NULL) == NULL) - { - ereport(ERROR, - (errmsg("tables hash table for " - "database %u corrupted during " - "cleanup --- abort", - dbentry->databaseid))); - } - } - continue; - } - - /* - * At least we think this is still a live table. Print its access - * stats. - */ fputc('T', fpout); fwrite(tabentry, sizeof(PgStat_StatTabEntry), 1, fpout); } @@ -2391,134 +2539,66 @@ pgstat_write_statsfile(void) fputc('d', fpout); } - /* - * Write out the known running backends to the stats file. - */ - i = MaxBackends; - fputc('M', fpout); - fwrite(&i, sizeof(i), 1, fpout); - - for (i = 0; i < MaxBackends; i++) - { - if (pgStatBeTable[i].procpid > 0) - { - fputc('B', fpout); - fwrite(&pgStatBeTable[i], sizeof(PgStat_StatBeEntry), 1, fpout); - } - } - /* * No more output to be done. Close the temp file and replace the old - * pgstat.stat with it. + * pgstat.stat with it. The ferror() check replaces testing for error + * after each individual fputc or fwrite above. */ fputc('E', fpout); - if (fclose(fpout) < 0) + + if (ferror(fpout)) { ereport(LOG, (errcode_for_file_access(), - errmsg("could not close temporary statistics file \"%s\": %m", + errmsg("could not write temporary statistics file \"%s\": %m", PGSTAT_STAT_TMPFILE))); + fclose(fpout); + unlink(PGSTAT_STAT_TMPFILE); } - else + else if (fclose(fpout) < 0) { - if (rename(PGSTAT_STAT_TMPFILE, PGSTAT_STAT_FILENAME) < 0) - { - ereport(LOG, - (errcode_for_file_access(), - errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m", - PGSTAT_STAT_TMPFILE, PGSTAT_STAT_FILENAME))); - } + ereport(LOG, + (errcode_for_file_access(), + errmsg("could not close temporary statistics file \"%s\": %m", + PGSTAT_STAT_TMPFILE))); + unlink(PGSTAT_STAT_TMPFILE); } - - /* - * Clear out the dead backends table - */ - hash_seq_init(&hstat, pgStatBeDead); - while ((deadbe = (PgStat_StatBeDead *) hash_seq_search(&hstat)) != NULL) + else if (rename(PGSTAT_STAT_TMPFILE, PGSTAT_STAT_FILENAME) < 0) { - /* - * Count down the destroy delay and remove entries where it reaches 0. - */ - if (--(deadbe->destroy) <= 0) - { - if (hash_search(pgStatBeDead, - (void *) &(deadbe->procpid), - HASH_REMOVE, NULL) == NULL) - { - ereport(ERROR, - (errmsg("dead-server-process hash table corrupted " - "during cleanup --- abort"))); - } - } + ereport(LOG, + (errcode_for_file_access(), + errmsg("could not rename temporary statistics file \"%s\" to \"%s\": %m", + PGSTAT_STAT_TMPFILE, PGSTAT_STAT_FILENAME))); + unlink(PGSTAT_STAT_TMPFILE); } } -/* - * qsort/bsearch comparison routine for PIDs - * - * We assume PIDs are nonnegative, so there's no overflow risk - */ -static int -comparePids(const void *v1, const void *v2) -{ - return *((const int *) v1) - *((const int *) v2); -} /* ---------- * pgstat_read_statsfile() - * - * Reads in an existing statistics collector and initializes the - * databases' hash table (whose entries point to the tables' hash tables) - * and the current backend table. + * Reads in an existing statistics collector file and initializes the + * databases' hash table (whose entries point to the tables' hash tables). * ---------- */ -static void -pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, - PgStat_StatBeEntry **betab, int *numbackends) +static HTAB * +pgstat_read_statsfile(Oid onlydb) { PgStat_StatDBEntry *dbentry; PgStat_StatDBEntry dbbuf; PgStat_StatTabEntry *tabentry; PgStat_StatTabEntry tabbuf; - PgStat_StatBeEntry *beentry; HASHCTL hash_ctl; + HTAB *dbhash; HTAB *tabhash = NULL; FILE *fpin; int32 format_id; - int maxbackends = 0; - int havebackends = 0; bool found; - int *live_pids; - MemoryContext use_mcxt; - int mcxt_flags; /* - * If running in the collector or the autovacuum process, we use the - * DynaHashCxt memory context. If running in a backend, we use the - * TopTransactionContext instead, so the caller must only know the last - * XactId when this call happened to know if his tables are still valid or - * already gone! - * - * Also, if running in a regular backend, we check backend entries against - * the PGPROC array so that we can detect stale entries. This lets us - * discard entries whose BETERM message got lost for some reason. + * The tables will live in pgStatLocalContext. */ - if (pgStatRunningInCollector || IsAutoVacuumProcess()) - { - use_mcxt = NULL; - mcxt_flags = 0; - live_pids = NULL; - } - else - { - use_mcxt = TopTransactionContext; - mcxt_flags = HASH_CONTEXT; - live_pids = GetAllBackendPids(); - /* Sort the PID array so we can use bsearch */ - if (live_pids[0] > 1) - qsort((void *) &live_pids[1], live_pids[0], sizeof(int), - comparePids); - } + pgstat_setup_memcxt(); /* * Create the DB hashtable @@ -2527,18 +2607,15 @@ pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, hash_ctl.keysize = sizeof(Oid); hash_ctl.entrysize = sizeof(PgStat_StatDBEntry); hash_ctl.hash = oid_hash; - hash_ctl.hcxt = use_mcxt; - *dbhash = hash_create("Databases hash", PGSTAT_DB_HASH_SIZE, &hash_ctl, - HASH_ELEM | HASH_FUNCTION | mcxt_flags); + hash_ctl.hcxt = pgStatLocalContext; + dbhash = hash_create("Databases hash", PGSTAT_DB_HASH_SIZE, &hash_ctl, + HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT); /* - * Initialize the number of known backends to zero, just in case we do a - * silent error return below. + * Clear out global statistics so they start from zero in case we can't + * load an existing statsfile. */ - if (numbackends != NULL) - *numbackends = 0; - if (betab != NULL) - *betab = NULL; + memset(&globalStats, 0, sizeof(globalStats)); /* * Try to open the status file. If it doesn't exist, the backends simply @@ -2546,7 +2623,7 @@ pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, * with empty counters. */ if ((fpin = AllocateFile(PGSTAT_STAT_FILENAME, PG_BINARY_R)) == NULL) - return; + return dbhash; /* * Verify it's of the expected format. @@ -2559,6 +2636,16 @@ pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, goto done; } + /* + * Read global stats struct + */ + if (fread(&globalStats, 1, sizeof(globalStats), fpin) != sizeof(globalStats)) + { + ereport(pgStatRunningInCollector ? LOG : WARNING, + (errmsg("corrupted pgstat.stat file"))); + goto done; + } + /* * We found an existing collector stats file. Read it and put all the * hashtable entries into place. @@ -2573,7 +2660,8 @@ pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, * until a 'd' is encountered. */ case 'D': - if (fread(&dbbuf, 1, sizeof(dbbuf), fpin) != sizeof(dbbuf)) + if (fread(&dbbuf, 1, offsetof(PgStat_StatDBEntry, tables), + fpin) != offsetof(PgStat_StatDBEntry, tables)) { ereport(pgStatRunningInCollector ? LOG : WARNING, (errmsg("corrupted pgstat.stat file"))); @@ -2583,7 +2671,7 @@ pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, /* * Add to the DB hash */ - dbentry = (PgStat_StatDBEntry *) hash_search(*dbhash, + dbentry = (PgStat_StatDBEntry *) hash_search(dbhash, (void *) &dbbuf.databaseid, HASH_ENTER, &found); @@ -2596,8 +2684,6 @@ pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, memcpy(dbentry, &dbbuf, sizeof(PgStat_StatDBEntry)); dbentry->tables = NULL; - dbentry->destroy = 0; - dbentry->n_backends = 0; /* * Don't collect tables if not the requested DB (or the @@ -2614,14 +2700,14 @@ pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, hash_ctl.keysize = sizeof(Oid); hash_ctl.entrysize = sizeof(PgStat_StatTabEntry); hash_ctl.hash = oid_hash; - hash_ctl.hcxt = use_mcxt; + hash_ctl.hcxt = pgStatLocalContext; dbentry->tables = hash_create("Per-database table", PGSTAT_TAB_HASH_SIZE, &hash_ctl, - HASH_ELEM | HASH_FUNCTION | mcxt_flags); + HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT); /* - * Arrange that following 'T's add entries to this databases + * Arrange that following 'T's add entries to this database's * tables hash table. */ tabhash = dbentry->tables; @@ -2638,7 +2724,8 @@ pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, * 'T' A PgStat_StatTabEntry follows. */ case 'T': - if (fread(&tabbuf, 1, sizeof(tabbuf), fpin) != sizeof(tabbuf)) + if (fread(&tabbuf, 1, sizeof(PgStat_StatTabEntry), + fpin) != sizeof(PgStat_StatTabEntry)) { ereport(pgStatRunningInCollector ? LOG : WARNING, (errmsg("corrupted pgstat.stat file"))); @@ -2660,100 +2747,9 @@ pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, ereport(pgStatRunningInCollector ? LOG : WARNING, (errmsg("corrupted pgstat.stat file"))); goto done; - } - - memcpy(tabentry, &tabbuf, sizeof(tabbuf)); - break; - - /* - * 'M' The maximum number of backends to expect follows. - */ - case 'M': - if (betab == NULL || numbackends == NULL) - goto done; - if (fread(&maxbackends, 1, sizeof(maxbackends), fpin) != - sizeof(maxbackends)) - { - ereport(pgStatRunningInCollector ? LOG : WARNING, - (errmsg("corrupted pgstat.stat file"))); - goto done; - } - if (maxbackends == 0) - goto done; - - /* - * Allocate space (in TopTransactionContext too) for the - * backend table. - */ - if (use_mcxt == NULL) - *betab = (PgStat_StatBeEntry *) - palloc(sizeof(PgStat_StatBeEntry) * maxbackends); - else - *betab = (PgStat_StatBeEntry *) - MemoryContextAlloc(use_mcxt, - sizeof(PgStat_StatBeEntry) * maxbackends); - break; - - /* - * 'B' A PgStat_StatBeEntry follows. - */ - case 'B': - if (betab == NULL || numbackends == NULL || *betab == NULL) - goto done; - - if (havebackends >= maxbackends) - goto done; - - /* - * Read it directly into the table. - */ - beentry = &(*betab)[havebackends]; - - if (fread(beentry, 1, sizeof(PgStat_StatBeEntry), fpin) != - sizeof(PgStat_StatBeEntry)) - { - ereport(pgStatRunningInCollector ? LOG : WARNING, - (errmsg("corrupted pgstat.stat file"))); - goto done; - } - - /* - * If possible, check PID to verify still running - */ - if (live_pids && - (live_pids[0] == 0 || - bsearch((void *) &beentry->procpid, - (void *) &live_pids[1], - live_pids[0], - sizeof(int), - comparePids) == NULL)) - { - /* - * Note: we could send a BETERM message to tell the - * collector to drop the entry, but I'm a bit worried - * about race conditions. For now, just silently ignore - * dead entries; they'll get recycled eventually anyway. - */ - - /* Don't accept the entry */ - memset(beentry, 0, sizeof(PgStat_StatBeEntry)); - break; - } - - /* - * Count backends per database here. - */ - dbentry = (PgStat_StatDBEntry *) - hash_search(*dbhash, - &(beentry->databaseid), - HASH_FIND, - NULL); - if (dbentry) - dbentry->n_backends++; - - havebackends++; - *numbackends = havebackends; + } + memcpy(tabentry, &tabbuf, sizeof(tabbuf)); break; /* @@ -2771,253 +2767,71 @@ pgstat_read_statsfile(HTAB **dbhash, Oid onlydb, done: FreeFile(fpin); + + return dbhash; } /* - * If not done for this transaction, read the statistics collector - * stats file into some hash tables. - * - * Because we store the hash tables in TopTransactionContext, the result - * is good for the entire current main transaction. - * - * Inside the autovacuum process, the statfile is assumed to be valid - * "forever", that is one iteration, within one database. This means - * we only consider the statistics as they were when the autovacuum - * iteration started. + * If not already done, read the statistics collector stats file into + * some hash tables. The results will be kept until pgstat_clear_snapshot() + * is called (typically, at end of transaction). */ static void backend_read_statsfile(void) { - if (IsAutoVacuumProcess()) - { - /* already read it? */ - if (pgStatDBHash) - return; - Assert(!pgStatRunningInCollector); - pgstat_read_statsfile(&pgStatDBHash, InvalidOid, - &pgStatBeTable, &pgStatNumBackends); - } - else - { - TransactionId topXid = GetTopTransactionId(); - - if (!TransactionIdEquals(pgStatDBHashXact, topXid)) - { - Assert(!pgStatRunningInCollector); - pgstat_read_statsfile(&pgStatDBHash, MyDatabaseId, - &pgStatBeTable, &pgStatNumBackends); - pgStatDBHashXact = topXid; - } - } -} - - -/* ---------- - * pgstat_recv_bestart() - - * - * Process a backend startup message. - * ---------- - */ -static void -pgstat_recv_bestart(PgStat_MsgBestart *msg, int len) -{ - PgStat_StatBeEntry *entry; - - /* - * If the backend is known dead, we ignore the message -- we don't want to - * update the backend entry's state since this BESTART message refers to - * an old, dead backend - */ - if (pgstat_add_backend(&msg->m_hdr) != 0) + /* already read it? */ + if (pgStatDBHash) return; + Assert(!pgStatRunningInCollector); - entry = &(pgStatBeTable[msg->m_hdr.m_backendid - 1]); - entry->userid = msg->m_userid; - memcpy(&entry->clientaddr, &msg->m_clientaddr, sizeof(entry->clientaddr)); - entry->databaseid = msg->m_databaseid; + /* Autovacuum launcher wants stats about all databases */ + if (IsAutoVacuumLauncherProcess()) + pgStatDBHash = pgstat_read_statsfile(InvalidOid); + else + pgStatDBHash = pgstat_read_statsfile(MyDatabaseId); } /* ---------- - * pgstat_recv_beterm() - - * - * Process a backend termination message. - * ---------- - */ -static void -pgstat_recv_beterm(PgStat_MsgBeterm *msg, int len) -{ - pgstat_sub_backend(msg->m_hdr.m_procpid); -} - -/* ---------- - * pgstat_recv_autovac() - + * pgstat_setup_memcxt() - * - * Process an autovacuum signalling message. + * Create pgStatLocalContext, if not already done. * ---------- */ static void -pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len) +pgstat_setup_memcxt(void) { - PgStat_StatDBEntry *dbentry; - - /* - * Lookup the database in the hashtable. Don't create the entry if it - * doesn't exist, because autovacuum may be processing a template - * database. If this isn't the case, the database is most likely to have - * an entry already. (If it doesn't, not much harm is done anyway -- - * it'll get created as soon as somebody actually uses the database.) - */ - dbentry = pgstat_get_db_entry(msg->m_databaseid, false); - if (dbentry == NULL) - return; - - /* - * Store the last autovacuum time in the database entry. - */ - dbentry->last_autovac_time = msg->m_start_time; + if (!pgStatLocalContext) + pgStatLocalContext = AllocSetContextCreate(TopMemoryContext, + "Statistics snapshot", + ALLOCSET_SMALL_MINSIZE, + ALLOCSET_SMALL_INITSIZE, + ALLOCSET_SMALL_MAXSIZE); } -/* ---------- - * pgstat_recv_vacuum() - - * - * Process a VACUUM message. - * ---------- - */ -static void -pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len) -{ - PgStat_StatDBEntry *dbentry; - PgStat_StatTabEntry *tabentry; - bool found; - bool create; - - /* - * If we don't know about the database, ignore the message, because it may - * be autovacuum processing a template database. But if the message is - * for database InvalidOid, don't ignore it, because we are getting a - * message from vacuuming a shared relation. - */ - create = (msg->m_databaseid == InvalidOid); - - dbentry = pgstat_get_db_entry(msg->m_databaseid, create); - if (dbentry == NULL) - return; - - tabentry = hash_search(dbentry->tables, &(msg->m_tableoid), - HASH_ENTER, &found); - - /* - * If we are creating the entry, initialize it. - */ - if (!found) - { - tabentry->numscans = 0; - - tabentry->tuples_returned = 0; - tabentry->tuples_fetched = 0; - tabentry->tuples_inserted = 0; - tabentry->tuples_updated = 0; - tabentry->tuples_deleted = 0; - - tabentry->n_live_tuples = msg->m_tuples; - tabentry->n_dead_tuples = 0; - - if (msg->m_analyze) - tabentry->last_anl_tuples = msg->m_tuples; - else - tabentry->last_anl_tuples = 0; - - tabentry->blocks_fetched = 0; - tabentry->blocks_hit = 0; - - tabentry->destroy = 0; - } - else - { - tabentry->n_live_tuples = msg->m_tuples; - tabentry->n_dead_tuples = 0; - if (msg->m_analyze) - tabentry->last_anl_tuples = msg->m_tuples; - } -} /* ---------- - * pgstat_recv_analyze() - + * pgstat_clear_snapshot() - * - * Process an ANALYZE message. - * ---------- - */ -static void -pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len) -{ - PgStat_StatDBEntry *dbentry; - PgStat_StatTabEntry *tabentry; - bool found; - - /* - * Note that we do create the database entry here, as opposed to what we - * do on AutovacStart and Vacuum messages. This is because autovacuum - * never executes ANALYZE on template databases. - */ - dbentry = pgstat_get_db_entry(msg->m_databaseid, true); - - tabentry = hash_search(dbentry->tables, &(msg->m_tableoid), - HASH_ENTER, &found); - - /* - * If we are creating the entry, initialize it. - */ - if (!found) - { - tabentry->numscans = 0; - - tabentry->tuples_returned = 0; - tabentry->tuples_fetched = 0; - tabentry->tuples_inserted = 0; - tabentry->tuples_updated = 0; - tabentry->tuples_deleted = 0; - - tabentry->n_live_tuples = msg->m_live_tuples; - tabentry->n_dead_tuples = msg->m_dead_tuples; - tabentry->last_anl_tuples = msg->m_live_tuples + msg->m_dead_tuples; - - tabentry->blocks_fetched = 0; - tabentry->blocks_hit = 0; - - tabentry->destroy = 0; - } - else - { - tabentry->n_live_tuples = msg->m_live_tuples; - tabentry->n_dead_tuples = msg->m_dead_tuples; - tabentry->last_anl_tuples = msg->m_live_tuples + msg->m_dead_tuples; - } -} - -/* ---------- - * pgstat_recv_activity() - + * Discard any data collected in the current transaction. Any subsequent + * request will cause new snapshots to be read. * - * Remember what the backend is doing. + * This is also invoked during transaction commit or abort to discard + * the no-longer-wanted snapshot. * ---------- */ -static void -pgstat_recv_activity(PgStat_MsgActivity *msg, int len) +void +pgstat_clear_snapshot(void) { - PgStat_StatBeEntry *entry; - - /* - * Here we check explicitly for 0 return, since we don't want to mangle - * the activity of an active backend by a delayed packet from a dead one. - */ - if (pgstat_add_backend(&msg->m_hdr) != 0) - return; - - entry = &(pgStatBeTable[msg->m_hdr.m_backendid - 1]); - - StrNCpy(entry->activity, msg->m_cmd_str, PGSTAT_ACTIVITY_SIZE); - - entry->activity_start_timestamp = GetCurrentTimestamp(); + /* Release memory, if any was allocated */ + if (pgStatLocalContext) + MemoryContextDelete(pgStatLocalContext); + + /* Reset variables */ + pgStatLocalContext = NULL; + pgStatDBHash = NULL; + localBackendStatusTable = NULL; + localNumBackends = 0; } @@ -3036,21 +2850,11 @@ pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len) int i; bool found; - /* - * Make sure the backend is counted for. - */ - if (pgstat_add_backend(&msg->m_hdr) < 0) - return; - dbentry = pgstat_get_db_entry(msg->m_databaseid, true); /* - * If the database is marked for destroy, this is a delayed UDP packet and - * not worth being counted. + * Update database-wide stats. */ - if (dbentry->destroy > 0) - return; - dbentry->n_xact_commit += (PgStat_Counter) (msg->m_xact_commit); dbentry->n_xact_rollback += (PgStat_Counter) (msg->m_xact_rollback); @@ -3069,48 +2873,50 @@ pgstat_recv_tabstat(PgStat_MsgTabstat *msg, int len) * If it's a new table entry, initialize counters to the values we * just got. */ - tabentry->numscans = tabmsg[i].t_numscans; - tabentry->tuples_returned = tabmsg[i].t_tuples_returned; - tabentry->tuples_fetched = tabmsg[i].t_tuples_fetched; - tabentry->tuples_inserted = tabmsg[i].t_tuples_inserted; - tabentry->tuples_updated = tabmsg[i].t_tuples_updated; - tabentry->tuples_deleted = tabmsg[i].t_tuples_deleted; - - tabentry->n_live_tuples = tabmsg[i].t_tuples_inserted; - tabentry->n_dead_tuples = tabmsg[i].t_tuples_updated + - tabmsg[i].t_tuples_deleted; - tabentry->last_anl_tuples = 0; - - tabentry->blocks_fetched = tabmsg[i].t_blocks_fetched; - tabentry->blocks_hit = tabmsg[i].t_blocks_hit; + tabentry->numscans = tabmsg[i].t_counts.t_numscans; + tabentry->tuples_returned = tabmsg[i].t_counts.t_tuples_returned; + tabentry->tuples_fetched = tabmsg[i].t_counts.t_tuples_fetched; + tabentry->tuples_inserted = tabmsg[i].t_counts.t_tuples_inserted; + tabentry->tuples_updated = tabmsg[i].t_counts.t_tuples_updated; + tabentry->tuples_deleted = tabmsg[i].t_counts.t_tuples_deleted; + tabentry->n_live_tuples = tabmsg[i].t_counts.t_new_live_tuples; + tabentry->n_dead_tuples = tabmsg[i].t_counts.t_new_dead_tuples; + tabentry->blocks_fetched = tabmsg[i].t_counts.t_blocks_fetched; + tabentry->blocks_hit = tabmsg[i].t_counts.t_blocks_hit; - tabentry->destroy = 0; + tabentry->last_anl_tuples = 0; + tabentry->vacuum_timestamp = 0; + tabentry->autovac_vacuum_timestamp = 0; + tabentry->analyze_timestamp = 0; + tabentry->autovac_analyze_timestamp = 0; } else { /* * Otherwise add the values to the existing entry. */ - tabentry->numscans += tabmsg[i].t_numscans; - tabentry->tuples_returned += tabmsg[i].t_tuples_returned; - tabentry->tuples_fetched += tabmsg[i].t_tuples_fetched; - tabentry->tuples_inserted += tabmsg[i].t_tuples_inserted; - tabentry->tuples_updated += tabmsg[i].t_tuples_updated; - tabentry->tuples_deleted += tabmsg[i].t_tuples_deleted; - - tabentry->n_live_tuples += tabmsg[i].t_tuples_inserted; - tabentry->n_dead_tuples += tabmsg[i].t_tuples_updated + - tabmsg[i].t_tuples_deleted; - - tabentry->blocks_fetched += tabmsg[i].t_blocks_fetched; - tabentry->blocks_hit += tabmsg[i].t_blocks_hit; + tabentry->numscans += tabmsg[i].t_counts.t_numscans; + tabentry->tuples_returned += tabmsg[i].t_counts.t_tuples_returned; + tabentry->tuples_fetched += tabmsg[i].t_counts.t_tuples_fetched; + tabentry->tuples_inserted += tabmsg[i].t_counts.t_tuples_inserted; + tabentry->tuples_updated += tabmsg[i].t_counts.t_tuples_updated; + tabentry->tuples_deleted += tabmsg[i].t_counts.t_tuples_deleted; + tabentry->n_live_tuples += tabmsg[i].t_counts.t_new_live_tuples; + tabentry->n_dead_tuples += tabmsg[i].t_counts.t_new_dead_tuples; + tabentry->blocks_fetched += tabmsg[i].t_counts.t_blocks_fetched; + tabentry->blocks_hit += tabmsg[i].t_counts.t_blocks_hit; } /* - * And add the block IO to the database entry. + * Add per-table stats to the per-database entry, too. */ - dbentry->n_blocks_fetched += tabmsg[i].t_blocks_fetched; - dbentry->n_blocks_hit += tabmsg[i].t_blocks_hit; + dbentry->n_tuples_returned += tabmsg[i].t_counts.t_tuples_returned; + dbentry->n_tuples_fetched += tabmsg[i].t_counts.t_tuples_fetched; + dbentry->n_tuples_inserted += tabmsg[i].t_counts.t_tuples_inserted; + dbentry->n_tuples_updated += tabmsg[i].t_counts.t_tuples_updated; + dbentry->n_tuples_deleted += tabmsg[i].t_counts.t_tuples_deleted; + dbentry->n_blocks_fetched += tabmsg[i].t_counts.t_blocks_fetched; + dbentry->n_blocks_hit += tabmsg[i].t_counts.t_blocks_hit; } } @@ -3125,15 +2931,8 @@ static void pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len) { PgStat_StatDBEntry *dbentry; - PgStat_StatTabEntry *tabentry; int i; - /* - * Make sure the backend is counted for. - */ - if (pgstat_add_backend(&msg->m_hdr) < 0) - return; - dbentry = pgstat_get_db_entry(msg->m_databaseid, false); /* @@ -3142,23 +2941,15 @@ pgstat_recv_tabpurge(PgStat_MsgTabpurge *msg, int len) if (!dbentry || !dbentry->tables) return; - /* - * If the database is marked for destroy, this is a delayed UDP packet and - * the tables will go away at DB destruction. - */ - if (dbentry->destroy > 0) - return; - /* * Process all table entries in the message. */ for (i = 0; i < msg->m_nentries; i++) { - tabentry = (PgStat_StatTabEntry *) hash_search(dbentry->tables, - (void *) &(msg->m_tableid[i]), - HASH_FIND, NULL); - if (tabentry) - tabentry->destroy = PGSTAT_DESTROY_COUNT; + /* Remove from hashtable if present; we don't care if it's not. */ + (void) hash_search(dbentry->tables, + (void *) &(msg->m_tableid[i]), + HASH_REMOVE, NULL); } } @@ -3174,22 +2965,26 @@ pgstat_recv_dropdb(PgStat_MsgDropdb *msg, int len) { PgStat_StatDBEntry *dbentry; - /* - * Make sure the backend is counted for. - */ - if (pgstat_add_backend(&msg->m_hdr) < 0) - return; - /* * Lookup the database in the hashtable. */ dbentry = pgstat_get_db_entry(msg->m_databaseid, false); /* - * Mark the database for destruction. + * If found, remove it. */ if (dbentry) - dbentry->destroy = PGSTAT_DESTROY_COUNT; + { + if (dbentry->tables != NULL) + hash_destroy(dbentry->tables); + + if (hash_search(pgStatDBHash, + (void *) &(dbentry->databaseid), + HASH_REMOVE, NULL) == NULL) + ereport(ERROR, + (errmsg("database hash table corrupted " + "during cleanup --- abort"))); + } } @@ -3205,12 +3000,6 @@ pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len) HASHCTL hash_ctl; PgStat_StatDBEntry *dbentry; - /* - * Make sure the backend is counted for. - */ - if (pgstat_add_backend(&msg->m_hdr) < 0) - return; - /* * Lookup the database in the hashtable. Nothing to do if not there. */ @@ -3231,7 +3020,6 @@ pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len) dbentry->n_xact_rollback = 0; dbentry->n_blocks_fetched = 0; dbentry->n_blocks_hit = 0; - dbentry->destroy = 0; memset(&hash_ctl, 0, sizeof(hash_ctl)); hash_ctl.keysize = sizeof(Oid); @@ -3242,3 +3030,133 @@ pgstat_recv_resetcounter(PgStat_MsgResetcounter *msg, int len) &hash_ctl, HASH_ELEM | HASH_FUNCTION); } + +/* ---------- + * pgstat_recv_autovac() - + * + * Process an autovacuum signalling message. + * ---------- + */ +static void +pgstat_recv_autovac(PgStat_MsgAutovacStart *msg, int len) +{ + PgStat_StatDBEntry *dbentry; + + /* + * Lookup the database in the hashtable. Don't create the entry if it + * doesn't exist, because autovacuum may be processing a template + * database. If this isn't the case, the database is most likely to have + * an entry already. (If it doesn't, not much harm is done anyway -- + * it'll get created as soon as somebody actually uses the database.) + */ + dbentry = pgstat_get_db_entry(msg->m_databaseid, false); + if (dbentry == NULL) + return; + + /* + * Store the last autovacuum time in the database entry. + */ + dbentry->last_autovac_time = msg->m_start_time; +} + +/* ---------- + * pgstat_recv_vacuum() - + * + * Process a VACUUM message. + * ---------- + */ +static void +pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len) +{ + PgStat_StatDBEntry *dbentry; + PgStat_StatTabEntry *tabentry; + + /* + * Don't create either the database or table entry if it doesn't already + * exist. This avoids bloating the stats with entries for stuff that is + * only touched by vacuum and not by live operations. + */ + dbentry = pgstat_get_db_entry(msg->m_databaseid, false); + if (dbentry == NULL) + return; + + tabentry = hash_search(dbentry->tables, &(msg->m_tableoid), + HASH_FIND, NULL); + if (tabentry == NULL) + return; + + if (msg->m_autovacuum) + tabentry->autovac_vacuum_timestamp = msg->m_vacuumtime; + else + tabentry->vacuum_timestamp = msg->m_vacuumtime; + tabentry->n_live_tuples = msg->m_tuples; + tabentry->n_dead_tuples = 0; + if (msg->m_analyze) + { + tabentry->last_anl_tuples = msg->m_tuples; + if (msg->m_autovacuum) + tabentry->autovac_analyze_timestamp = msg->m_vacuumtime; + else + tabentry->analyze_timestamp = msg->m_vacuumtime; + } + else + { + /* last_anl_tuples must never exceed n_live_tuples */ + tabentry->last_anl_tuples = Min(tabentry->last_anl_tuples, + msg->m_tuples); + } +} + +/* ---------- + * pgstat_recv_analyze() - + * + * Process an ANALYZE message. + * ---------- + */ +static void +pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len) +{ + PgStat_StatDBEntry *dbentry; + PgStat_StatTabEntry *tabentry; + + /* + * Don't create either the database or table entry if it doesn't already + * exist. This avoids bloating the stats with entries for stuff that is + * only touched by analyze and not by live operations. + */ + dbentry = pgstat_get_db_entry(msg->m_databaseid, false); + if (dbentry == NULL) + return; + + tabentry = hash_search(dbentry->tables, &(msg->m_tableoid), + HASH_FIND, NULL); + if (tabentry == NULL) + return; + + if (msg->m_autovacuum) + tabentry->autovac_analyze_timestamp = msg->m_analyzetime; + else + tabentry->analyze_timestamp = msg->m_analyzetime; + tabentry->n_live_tuples = msg->m_live_tuples; + tabentry->n_dead_tuples = msg->m_dead_tuples; + tabentry->last_anl_tuples = msg->m_live_tuples + msg->m_dead_tuples; +} + + +/* ---------- + * pgstat_recv_bgwriter() - + * + * Process a BGWRITER message. + * ---------- + */ +static void +pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len) +{ + globalStats.timed_checkpoints += msg->m_timed_checkpoints; + globalStats.requested_checkpoints += msg->m_requested_checkpoints; + globalStats.buf_written_checkpoints += msg->m_buf_written_checkpoints; + globalStats.buf_written_lru += msg->m_buf_written_lru; + globalStats.buf_written_all += msg->m_buf_written_all; + globalStats.maxwritten_lru += msg->m_maxwritten_lru; + globalStats.maxwritten_all += msg->m_maxwritten_all; +}