Percona Server with XtraDB

Patch to remove excessive fcntl() calls was never ported correctly to 5.1

Reported by Laurynas Biveinis on 2012-10-03
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
Percona Server
Undecided
Unassigned
5.1
Medium
Hrvoje Matijakovic
5.5
Undecided
Unassigned

Bug Description

This patch -
http://www.percona.com/doc/percona-server/5.1/performance/remove_fcntl_excessive_calls.html
https://bugs.launchpad.net/percona-server/+bug/606810

has always been a no-op in PS 5.1.

The reason is that the patch works by preprocessor, adding "... && !defined(NO_ALARM) " several times to net_serv.cc to disable the alarm code. But nothing in PS build process or official binary build scripts actually define NO_ALARM, it never gets defined, the alarm code never gets disabled.

Not a bug on PS 5.5, as CMake build sys starts defining NO_ALARM for us.

Related branches

lp:~hrvojem/percona-server/rn-5.1.66-14.1-5.1
Merged into lp:percona-server/5.1 at revision 494
Laurynas Biveinis: Needs Fixing on 2012-10-26
Stewart Smith (community): Approve on 2012-10-23
Alexey Kopytov (akopytov) wrote :

I'm more in favor of keeping the status quo and just fixing the docs to remove the feature, because:

1. Even though the patch was there, the feature was never enabled in our official 5.1 binaries
2. Enabling the feature in our 5.1 binaries would lead to either http://bugs.mysql.com/bug.php?id=52633, or (if the fix is backported) to an unconditional change from the documented behavior in net_read_timeout/net_write_timeout/net_retry_count in upstream 5.1.

Either outcome doesn't look good to me. Let's change it to a doc bug?

Yes, and my MP'ed fix already includes the 52633 backport. Can we document that net_read_timeout and friends behave like on 5.5 instead? IM*H*O it's a relatively low-impact changes that comes close to DTRT.

One more thing backported by this fix is that SIGNAL_WITH_VIO_CLOSE is unconditionally defined, just like on 5.5.

Alexey Kopytov (akopytov) wrote :

I don't like deviating from a documented 5.1 upstream behavior. There must be a way to move from and to upstream without changes to configuration file (and enabling PS-specific features).

Alexey Kopytov (akopytov) wrote :

And another argument for keeping the status quo in 5.1 is that it's legacy. I.e. "don't touch unless you absolutely have to".

Right, PS is not a drop-in anymore. Re-assigning to Hrvoje to document the current state.

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.