From: Justin Pryzby <pryzby@telsasoft.com>
To: Julien Rouhaud <rjuju123@gmail.com>
To: Michael Paquier <michael@paquier.xyz>
Cc: Sergei Kornilov <sk@zsrv.org>
Cc: Guillaume Lelarge <guillaume@lelarge.info>
Cc: pgsql-hackers@lists.postgresql.org
Subject: Re: Expose lock group leader pid in pg_stat_activity
Date: Sun, 15 Mar 2020 23:27:52 -0500
Message-ID: <20200316042752.GE26184@telsasoft.com> (raw)
In-Reply-To: <20200116074912.GA98418@paquier.xyz> <CAOBaU_Z6WcT8FJfcCMU_Bp+PTzUpTEY9BfBQE4gbGKpE5wpfwQ@mail.gmail.com>
On Tue, Jan 28, 2020 at 12:36:41PM +0100, Julien Rouhaud wrote:
> So, AFAICT the LockHashPartitionLockByProc is required when
> iterating/modifying lockGroupMembers or lockGroupLink, but just
> getting the leader pid should be safe.
This still seems unsafe:
git show -U11 -w --patience b025f32e0b src/backend/utils/adt/pgstatfuncs.c
+ /*
+ * If a PGPROC entry was retrieved, display wait events and lock
+ * group leader information if any. To avoid extra overhead, no
+ * extra lock is being held, so there is no guarantee of
+ * consistency across multiple rows.
+ */
...
+ PGPROC *leader;
...
+ leader = proc->lockGroupLeader;
+ if (leader)
+ {
# does something guarantee that leader doesn't change ?
+ values[29] = Int32GetDatum(leader->pid);
+ nulls[29] = false;
}
Michael seems to have raised the issue:
On Thu, Jan 16, 2020 at 04:49:12PM +0900, Michael Paquier wrote:
> And actually, the way you are looking at the leader's PID is visibly
> incorrect and inconsistent because the patch takes no shared LWLock on
> the leader using LockHashPartitionLockByProc() followed by
> LWLockAcquire(), no? That's incorrect because it could be perfectly
> possible to crash with this code between the moment you check if
> lockGroupLeader is NULL and the moment you look at
> lockGroupLeader->pid if a process is being stopped in-between and
> removes itself from a lock group in ProcKill().
But I don't see how it was addressed ?
I read this:
src/backend/storage/lmgr/lock.c: * completely valid. We cannot safely dereference another backend's
src/backend/storage/lmgr/lock.c- * lockGroupLeader field without holding all lock partition locks, and
src/backend/storage/lmgr/lock.c- * it's not worth that.)
I think if you do:
|LWLockAcquire(&proc->backendLock, LW_SHARED);
..then it's at least *safe* to access leader->pid, but it may be inconsistent
unless you also call LockHashPartitionLockByProc.
I wasn't able to produce a crash, so maybe I missed something.
--
Justin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-hackers@postgresql.org
Cc: pryzby@telsasoft.com, rjuju123@gmail.com, michael@paquier.xyz, sk@zsrv.org, guillaume@lelarge.info, pgsql-hackers@lists.postgresql.org
Subject: Re: Expose lock group leader pid in pg_stat_activity
In-Reply-To: <20200316042752.GE26184@telsasoft.com>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox