database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-11-16 15:02:33 +03:00
Code Activity

Handle impending sinval queue overflow by means of a separate signal

Browse Source 
(SIGUSR1, which we have not been using recently) instead of piggybacking
on SIGUSR2-driven NOTIFY processing.  This has several good results:
the processing needed to drain the sinval queue is a lot less than the
processing needed to answer a NOTIFY; there's less contention since we
don't have a bunch of backends all trying to acquire exclusive lock on
pg_listener; backends that are sitting inside a transaction block can
still drain the queue, whereas NOTIFY processing can't run if there's
an open transaction block.  (This last is a fairly serious issue that
I don't think we ever recognized before --- with clients like JDBC that
tend to sit with open transaction blocks, the sinval queue draining
mechanism never really worked as intended, probably resulting in a lot
of useless cache-reset overhead.)  This is the last of several proposed
changes in response to Philip Warner's recent report of sinval-induced
performance problems.
This commit is contained in:
Tom Lane
2004-05-23 03:50:45 +00:00
parent 4d86ae4260
commit ebfc56d3fb
8 changed files with 289 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/backend/storage/ipc/sinvaladt.c
View File

@@ -8,7 +8,7 @@
*
*
* IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/storage/ipc/sinvaladt.c,v 1.54 2003/12/20 17:31:21 momjian Exp $
* $PostgreSQL: pgsql/src/backend/storage/ipc/sinvaladt.c,v 1.55 2004/05/23 03:50:45 tgl Exp $
*
*-------------------------------------------------------------------------
*/
@@ -215,16 +215,12 @@ SIInsertDataEntry(SISeg *segP, SharedInvalidationMessage *data)
/*
* Try to prevent table overflow. When the table is 70% full send a
* WAKEN_CHILDREN request to the postmaster. The postmaster will send
* a SIGUSR2 signal (ordinarily a NOTIFY signal) to all the backends.
* This will force idle backends to execute a transaction to look
* through pg_listener for NOTIFY messages, and as a byproduct of the
* transaction start they will read SI entries.
* a SIGUSR1 signal to all the backends, which will cause sinval.c
* to read any pending SI entries.
*
* This should never happen if all the backends are actively executing
* queries, but if a backend is sitting idle then it won't be starting
* transactions and so won't be reading SI entries.
*
* dz - 27 Jan 1998
*/
if (numMsgs == (MAXNUMMESSAGES * 70 / 100) &&
IsUnderPostmaster)