[Zope] Re: Leaking Oracle connections processes
Dieter Maurer wrote: Shai Berger writes: ... excess Oracle processes and zombies ... I have checked our installation and can confirm that both problems have gone. We do not yet know what caused the excess Oracle processes. After an increase of Zope's ZODB cache and an upgrade to the newest DCOracle, the problem simply went away. Do you have a DCOracle version later than 1.3.2? We use 1.3.2; we have increased the cache from 400 (the default) to 1000, to no avail. Our ZODB size, last I checked, was ~40M. I remember you mentioning that yours was less then 10M, I think; how much cache do you use? We do know what caused our zombie processes: a bug in the Oracle Java runtime library: they had added a "NOWAIT" flag to their "wait" calls. This ensured that a "wait" call does not release the Zombie child but keeps it in its state. As a consequence, each process started from Java became a Zombie process. Zope starts Oracle processes from Java??? All the zombies we see are Zope-spawned... a Zope zombie-collection problem has been reported independently of Oracle. As Oracle reported, the problem was introduced in 8.1.6 and should be fixed in 8.1.7. ...which confuses me, because we use 8.0.5. Thanks, Shai. ___ Zope maillist - [EMAIL PROTECTED] http://lists.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://lists.zope.org/mailman/listinfo/zope-announce http://lists.zope.org/mailman/listinfo/zope-dev )
[Zope] Re: Leaking Oracle connections processes
Shai Berger wrote: I have checked a little before the disaster, and it seems that not all those processes actually have connections; the problem may be unrelated to open connections, but just to processes. A little progress: We just had it happen again, and checked to see how many live processes and how many zombies were involved. pstree showed 108 Zope-spawned Oracle processes, but ps showed only 55 zombies. I assume, therefore, that the problem is not because of zombies; there were over 50 live processes accessing oracle. The accumulation of zombies is a problem in itself, of course, but I understand it is also a known problem. And from what I see, it is a separate problem. Our impression here -- based on not much more than gut feeling -- is that the problem is related to abandoned transactions, where an exception is raised. Thanks, Dieter, for your earlier response -- we are trying a larger cache now, and we'll see how things go. Have fun, Shai. ___ Zope maillist - [EMAIL PROTECTED] http://lists.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://lists.zope.org/mailman/listinfo/zope-announce http://lists.zope.org/mailman/listinfo/zope-dev )
[Zope] Re: Leaking Oracle connections processes
Shai Berger writes: ... excess Oracle processes and zombies ... I have checked our installation and can confirm that both problems have gone. We do not yet know what caused the excess Oracle processes. After an increase of Zope's ZODB cache and an upgrade to the newest DCOracle, the problem simply went away. We do know what caused our zombie processes: a bug in the Oracle Java runtime library: they had added a "NOWAIT" flag to their "wait" calls. This ensured that a "wait" call does not release the Zombie child but keeps it in its state. As a consequence, each process started from Java became a Zombie process. As Oracle reported, the problem was introduced in 8.1.6 and should be fixed in 8.1.7. Dieter ___ Zope maillist - [EMAIL PROTECTED] http://lists.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://lists.zope.org/mailman/listinfo/zope-announce http://lists.zope.org/mailman/listinfo/zope-dev )