Re: Should PostgresMain() do a LWLockReleaseAll()? - Mailing list pgsql-hackers

From Andres Freund
Subject Re: Should PostgresMain() do a LWLockReleaseAll()?
Date
Msg-id 20140223202938.GA20412@awork2.anarazel.de
Whole thread Raw
In response to Re: Should PostgresMain() do a LWLockReleaseAll()?  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-hackers
On 2014-02-23 14:48:12 -0500, Tom Lane wrote:
> Andres Freund <andres@2ndquadrant.com> writes:
> > Currently the error handling of normal backends only does a
> > LWLockReleaseAll() once CurrentTransactionState->state != TRANS_DEFAULT
> > because it's called in AbortTransaction(). There's pretty damn few
> > places that fiddle with lwlocks outside of a transaction command, but I
> > still do wonder whether it'd wouldn't be a tad more robust to
> > unconditionally do a LWLockReleaseAll(), just like other error handlers
> > are doing?
>
> Why do that thing in particular, and not all the other things that
> AbortTransaction() does?

Because the other things in AbortTransaction() should really only be
relevant inside a transaction, but there's valid reasons to use lwlocks
outside one.

E.g. I think that before Robert and I added a LWLockReleaseAll() to
WalSndErrorCleanup() the whole walsender code wasn't protected. I am not
entirely sure there's a real problem there in the backbranches, but it's
a fair amount of code, espcially around base backups...

Greetings,

Andres Freund

--Andres Freund                       http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services



pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Re: [BUGS] Re: Re: BUG #9210: PostgreSQL string store bug? not enforce check with correct characterSET/encoding
Next
From: Andres Freund
Date:
Subject: Re: often PREPARE can generate high load (and sometimes minutes long unavailability)