From: Badlop <badlop@process-one.net>
Date: Fri, 13 Feb 2009 19:27:56 +0000 (+0000)
Subject: * src/ejabberd_system_monitor.erl: Allow parametrizable watchdog
X-Git-Tag: v2.1.0~18^2~446
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=3524172856fea6cc5be3e0987f1acb3fd1524d0b;p=ejabberd

* src/ejabberd_system_monitor.erl: Allow parametrizable watchdog
threshold: option watchdog_large_heap or chatting with the
watchdog bot (EJAB-545)
* src/ejabberd_config.erl: Likewise
* doc/guide.tex: Likewise
* doc/guide.html: Likewise

SVN Revision: 1872
---

diff --git a/ChangeLog b/ChangeLog
index 4d962b827..baff417c6 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,12 @@
+2009-02-13  Badlop  <badlop@process-one.net>
+
+	* src/ejabberd_system_monitor.erl: Allow parametrizable watchdog
+	threshold: option watchdog_large_heap or chatting with the
+	watchdog bot (EJAB-545)
+	* src/ejabberd_config.erl: Likewise
+	* doc/guide.tex: Likewise
+	* doc/guide.html: Likewise
+
 2009-02-13  Christophe Romain <christophe.romain@process-one.net>
 
 	* src/ejabberd_auth.erl: prevent from calling
diff --git a/doc/guide.html b/doc/guide.html
index 645ae6d5c..55ce1dc69 100644
--- a/doc/guide.html
+++ b/doc/guide.html
@@ -3366,14 +3366,16 @@ There are some simple and safe examples in the article
 <H2 CLASS="section"><!--SEC ANCHOR --><A NAME="htoc87">7.3</A>&#XA0;&#XA0;<A HREF="#watchdog">Watchdog Alerts</A></H2><!--SEC END --><P> <A NAME="watchdog"></A> 
 </P><P><TT>ejabberd</TT> includes a watchdog mechanism that may be useful to developers
 when troubleshooting a problem related to memory usage.
-If a process in the <TT>ejabberd</TT> server consumes a lot of memory,
+If a process in the <TT>ejabberd</TT> server consumes more memory than the configured threshold,
 a message is sent to the Jabber accounts defined with the option
 <TT>watchdog_admins</TT>
- in the <TT>ejabberd</TT> configuration file.
-Note that the threshold to define what is too much memory usage 
-is only configurable editing the source code.
-Example configuration:
+ in the <TT>ejabberd</TT> configuration file.</P><P>The memory consumed is measured in <TT>words</TT>:
+a word on 32-bit architecture is 4 bytes, and a word on 64-bit architecture 8 bytes.
+The threshold by default is 1000000 words.
+This value can be configured with the option <TT>watchdog_large_heap</TT>,
+or in a conversation with the watchdog alert bot.</P><P>Example configuration:
 </P><PRE CLASS="verbatim">{watchdog_admins, ["admin2@localhost", "admin2@example.org"]}.
+{watchdog_large_heap, 30000000}.
 </PRE><P>To remove watchdog admins, remove them in the option.
 To remove all watchdog admins, set the option with an empty list:
 </P><PRE CLASS="verbatim">{watchdog_admins, []}.
diff --git a/doc/guide.tex b/doc/guide.tex
index e7ea7c13e..169dc612e 100644
--- a/doc/guide.tex
+++ b/doc/guide.tex
@@ -4428,15 +4428,22 @@ To exit the shell, close the window or press the keys: control+c control+c.
 
 \ejabberd{} includes a watchdog mechanism that may be useful to developers
 when troubleshooting a problem related to memory usage.
-If a process in the \ejabberd{} server consumes a lot of memory,
+If a process in the \ejabberd{} server consumes more memory than the configured threshold,
 a message is sent to the Jabber accounts defined with the option
 \term{watchdog\_admins}
 \ind{options!watchdog\_admins} in the \ejabberd{} configuration file.
-Note that the threshold to define what is too much memory usage 
-is only configurable editing the source code.
+
+The memory consumed is measured in \term{words}:
+a word on 32-bit architecture is 4 bytes,
+and a word on 64-bit architecture is 8 bytes.
+The threshold by default is 1000000 words.
+This value can be configured with the option \term{watchdog\_large\_heap},
+or in a conversation with the watchdog alert bot.
+
 Example configuration:
 \begin{verbatim}
 {watchdog_admins, ["admin2@localhost", "admin2@example.org"]}.
+{watchdog_large_heap, 30000000}.
 \end{verbatim}
 
 To remove watchdog admins, remove them in the option.
@@ -4445,7 +4452,6 @@ To remove all watchdog admins, set the option with an empty list:
 {watchdog_admins, []}.
 \end{verbatim}
 
-
 \appendix{}
 
 \makechapter{i18ni10n}{Internationalization and Localization}
diff --git a/src/ejabberd_config.erl b/src/ejabberd_config.erl
index ecf8ec30f..dc54ee884 100644
--- a/src/ejabberd_config.erl
+++ b/src/ejabberd_config.erl
@@ -356,6 +356,8 @@ process_term(Term, State) ->
 	    add_option({domain_balancing_component_number, Domain}, N, State);
 	{watchdog_admins, Admins} ->
 	    add_option(watchdog_admins, Admins, State);
+	{watchdog_large_heap, LH} ->
+	    add_option(watchdog_large_heap, LH, State);
 	{registration_timeout, Timeout} ->
 	    add_option(registration_timeout, Timeout, State);
 	{loglevel, Loglevel} ->
diff --git a/src/ejabberd_system_monitor.erl b/src/ejabberd_system_monitor.erl
index 5fa2cc4d9..505b0994b 100644
--- a/src/ejabberd_system_monitor.erl
+++ b/src/ejabberd_system_monitor.erl
@@ -51,7 +51,12 @@
 %% Description: Starts the server
 %%--------------------------------------------------------------------
 start_link() ->
-    gen_server:start_link({local, ?MODULE}, ?MODULE, [], []).
+    LH = case ejabberd_config:get_local_option(watchdog_large_heap) of
+	I when is_integer(I) -> I;
+	_ -> 1000000
+end,
+    Opts = [{large_heap, LH}],
+    gen_server:start_link({local, ?MODULE}, ?MODULE, Opts, []).
 
 process_command(From, To, Packet) ->
     case To of
@@ -90,9 +95,10 @@ process_command(From, To, Packet) ->
 %%                         {stop, Reason}
 %% Description: Initiates the server
 %%--------------------------------------------------------------------
-init([]) ->
+init(Opts) ->
+    LH = proplists:get_value(large_heap, Opts),
     process_flag(priority, high),
-    erlang:system_monitor(self(), [{large_heap, 1000000}]),
+    erlang:system_monitor(self(), [{large_heap, LH}]),
     lists:foreach(
       fun(Host) ->
 	      ejabberd_hooks:add(local_send_to_resource_hook, Host,
@@ -109,10 +115,24 @@ init([]) ->
 %%                                      {stop, Reason, State}
 %% Description: Handling call messages
 %%--------------------------------------------------------------------
+handle_call({get, large_heap}, _From, State) ->
+    {reply, get_large_heap(), State};
+handle_call({set, large_heap, NewValue}, _From, State) ->
+    MonSettings = erlang:system_monitor(self(), [{large_heap, NewValue}]),
+    OldLH = get_large_heap(MonSettings),
+    NewLH = get_large_heap(),
+    {reply, {lh_changed, OldLH, NewLH}, State};
 handle_call(_Request, _From, State) ->
     Reply = ok,
     {reply, Reply, State}.
 
+get_large_heap() ->
+    MonSettings = erlang:system_monitor(),
+    get_large_heap(MonSettings).
+get_large_heap(MonSettings) ->
+    {_MonitorPid, Options} = MonSettings,
+    proplists:get_value(large_heap, Options).
+
 %%--------------------------------------------------------------------
 %% Function: handle_cast(Msg, State) -> {noreply, State} |
 %%                                      {noreply, State, Timeout} |
@@ -165,7 +185,7 @@ process_large_heap(Pid, Info) ->
 		  JIDs /= [] ->
 	    DetailedInfo = detailed_info(Pid),
 	    Body = io_lib:format(
-		     "(~w) The process ~w is consuming too much memory: ~w.~n"
+		     "(~w) The process ~w is consuming too much memory:~n~p~n"
 		     "~s",
 		     [node(), Pid, Info, DetailedInfo]),
 	    From = jlib:make_jid("", Host, "watchdog"),
@@ -303,14 +323,25 @@ process_command1(From, To, Body) ->
 process_command2(["kill", SNode, SPid], From, To) ->
     Node = list_to_atom(SNode),
     remote_command(Node, [kill, SPid], From, To);
+process_command2(["showlh", SNode], From, To) ->
+    Node = list_to_atom(SNode),
+    remote_command(Node, [showlh], From, To);
+process_command2(["setlh", SNode, NewValueString], From, To) ->
+    Node = list_to_atom(SNode),
+    NewValue = list_to_integer(NewValueString),
+    remote_command(Node, [setlh, NewValue], From, To);
 process_command2(["help"], From, To) ->
     send_message(To, From, help());
 process_command2(_, From, To) ->
     send_message(To, From, help()).
 
+
 help() ->
     "Commands:\n"
-	"  kill <node> <pid>".
+	"  kill <node> <pid>\n"
+	"  showlh <node>\n"
+	"  setlh <node> <integer>".
+
 
 remote_command(Node, Args, From, To) ->
     Message =
@@ -325,6 +356,12 @@ remote_command(Node, Args, From, To) ->
 process_remote_command([kill, SPid]) ->
     exit(list_to_pid(SPid), kill),
     "ok";
+process_remote_command([showlh]) ->
+    Res = gen_server:call(ejabberd_system_monitor, {get, large_heap}),
+    io_lib:format("Current large heap: ~p", [Res]);
+process_remote_command([setlh, NewValue]) ->
+    {lh_changed, OldLH, NewLH} = gen_server:call(ejabberd_system_monitor, {set, large_heap, NewValue}),
+    io_lib:format("Result of set large heap: ~p --> ~p", [OldLH, NewLH]);
 process_remote_command(_) ->
     throw(unknown_command).