Re: Patch to allow users to kill their own queries - Mailing list pgsql-hackers

From Greg Smith
Subject Re: Patch to allow users to kill their own queries
Date
Msg-id 4EEB3A2B.8090102@2ndQuadrant.com
Whole thread Raw
In response to Re: Patch to allow users to kill their own queries  (Magnus Hagander <magnus@hagander.net>)
Responses Re: Patch to allow users to kill their own queries  (Magnus Hagander <magnus@hagander.net>)
List pgsql-hackers
On 12/14/2011 05:24 AM, Magnus Hagander wrote:
> How about passing a parameter to pg_signal_backend? Making
> pg_signal_backend(int pid, int sig, bool allow_samerole)?
>

That works, got rid of the parts I didn't like and allowed some useful
minor restructuring.  I also made the HINT better and match style
guidelines:

gsmith=> select pg_terminate_backend(21205);
ERROR:  must be superuser to terminate other server processes
HINT:  You can cancel your own processes with pg_cancel_backend().
gsmith=> select pg_cancel_backend(21205);
  pg_cancel_backend
-------------------
  t

New rev attached and pushed to
https://github.com/greg2ndQuadrant/postgres/tree/cancel-backend (which
is *not* the same branch as I used last time; don't ask why, long story)

I considered some additional ways to restructure the checks that could
remove a further line or two from the logic here, but they all made the
result seem less readable to me.  And this is not a high performance
code path.  I may have gone a bit too far with the comment additions
though, so feel free to trim that back.  It kept feeling weird to me
that none of the individual signaling functions had their own intro
comments.  I added all those.

I also wrote up a commentary on the PID wraparound race condition
possibility Josh brought up.  Some research shows that pid assignment on
some systems is made more secure by assigning new ones randomly.  That
seems like it would make it possible to have a pid get reused much
faster than on the usual sort of system that does sequential assignment
and wraparound.  A reuse collision still seems extremely unlikely
though.  With the new comments, at least a future someone who speculates
on this will know how much thinking went into the current
implementation:  enough to notice, not enough to see anything worth
doing about it.  Maybe that's just wasted lines of text?

With so little grief on the last round, I'm going to guess this one will
just get picked up by Magnus to commit next.  Marking accordingly and
moved to the current CommitFest.

--
Greg Smith   2ndQuadrant US    greg@2ndQuadrant.com   Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support  www.2ndQuadrant.us


Attachment

pgsql-hackers by date:

Previous
From: Simon Riggs
Date:
Subject: Re: ALTER TABLE lock strength reduction patch is unsafe
Next
From: Simon Riggs
Date:
Subject: Re: Moving more work outside WALInsertLock