[EMAIL PROTECTED] wrote: > api.py: > > We shouldn't be exposing use_cache to API consumers. Shawn and I have > both said that the API should be modified to disable caching for > operations that don't modify the image. This change disables it > everywhere all the time, and exposes that feature to the API > I agree that would be the sensible thing to do, but the problem is that if we do that in the PM at the minute then we still hit the memory consumption issue as a user carries out several operations against the same api object and hence the same image. The CLI carries out one predefined action and then exits so doesn't suffer from this to the same extent.
I'd love to get to the bottom of this so we can put in the changes you suggest above, but it will take us to dig deeper into python and perhaps get the python.org core developers involved if we can give them a simple reproducible test case. > consumers. > > image.py: > > As stated previously, it seems like this should be done on a > per-operation basis, not a per-image basis. > > Since this is a workaround, expect that you'll file a bug to fix the > real problem as well as a bug to remove the workaround. Please make the > bug to remove the workaround depend upon the bug to fix the problem. > > Very much agree that this is a temporary workaround for 2008.11 and once we figure out what is going on we should fix the real issue and remove this workaround. Thanks, JR > Thanks. > > -j > > On Fri, Oct 24, 2008 at 09:18:02AM +0100, jmr wrote: > >> Michal has respun his webrev without the actions:__init__.py revert to r590 >> as well, to give us both options. >> >> With __init__.py from the trunk: >> http://cr.opensolaris.org/~migi/23_10_2008_bug_4116_v1_no_init/ >> >> With __init__.py from rev590: >> http://cr.opensolaris.org/~migi/23_10_2008_bug_4116_v1/ >> >> Thanks Michal. >> >> JR >> >> jmr wrote: >> >>> This webrev of actually rolls back Michals actions:__init__.py to r590. >>> I'm not sure if this is what Danek wants us to do, or if he wants to tweak >>> the code himself. Reverting is just a workaround and does not get to the >>> root cause of the memory leak, but given the time scales this may be our >>> only option at present. >>> >>> Danek, do you want this revert of actions:__init__.py to r590 to happen or >>> not? >>> >>> Michal can you respin this webrev without the reversion of >>> actions:__init__.py to r590 so we have two options available. >>> >>> Thanks. >>> >>> JR >>> >>> jmr wrote: >>> >>> >>>> Hi - these changes by Michal have been under discussion over the past >>>> week on the bug 4116 and reviewed by Brock and myself several times. This >>>> webrev addresses the issue from the PM and UM side of things by making >>>> various optimizations, including turning off the manifest cache only for >>>> PM and UM, and using Brock's new actions support in the api to further >>>> reduce the need to access the manifest directly. >>>> >>>> http://cr.opensolaris.org/~migi/23_10_2008_bug_4116_v1/ >>>> <http://cr.opensolaris.org/%7Emigi/23_10_2008_bug_4116_v1/> >>>> >>>> All the details including analysis are in the bug: >>>> http://defect.opensolaris.org/bz/show_bug.cgi?id=4116 >>>> >>>> Running PM with the gate: >>>> Mem: 600M >>>> >>>> Running PM with this webrev alone against gate: >>>> Mem: 480M >>>> >>>> Running PM with this webrev and actions:__init__.py at rev590 against gate: >>>> Mem: 180M >>>> >>>> There is still the outstanding issue of a memory increase caused by >>>> introduction of changes going from r590 to r591 to actions:__inti__.py >>>> that Danek is investigating. This is a stopper bug for PM and the 2008.11 >>>> release. >>>> >>>> JR >>>> >>>> >>>> <http://cr.opensolaris.org/%7Emigi/23_10_2008_bug_4116_v1/> >>>> >>>> _______________________________________________ >>>> pkg-discuss mailing list >>>> [email protected] >>>> http://mail.opensolaris.org/mailman/listinfo/pkg-discuss >>>> >>>> >>> _______________________________________________ >>> pkg-discuss mailing list >>> [email protected] >>> http://mail.opensolaris.org/mailman/listinfo/pkg-discuss >>> >>> > _______________________________________________ > pkg-discuss mailing list > [email protected] > http://mail.opensolaris.org/mailman/listinfo/pkg-discuss > _______________________________________________ pkg-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
