| 00:00.31 | *** join/#storm wallflower (n=wallflow@ip-205-246-113-216.pool.grokthis.net) |
| 00:05.28 | *** join/#storm thumper (n=tim@125-236-193-95.adsl.xtra.co.nz) |
| 00:19.29 | *** join/#storm thumper_laptop (n=tim@125-236-193-95.adsl.xtra.co.nz) |
| 03:45.53 | *** join/#storm bigdog (n=scmikes@72-197-8-8-arpa.cust.cinci.current.net) |
| 04:15.52 | *** join/#storm nullie (n=ilya@87.254.139.66) |
| 04:25.59 | *** join/#storm jukart (n=jukart@85-124-221-45.static.xdsl-line.inode.at) |
| 05:16.50 | *** join/#storm nullie (n=ilya@87.254.139.66) |
| 06:50.16 | *** join/#storm jukart (n=jukart@lsfw01.lovelysystems.com) |
| 07:04.37 | *** join/#storm mcella (n=mcella@ip-177-201.sn2.eutelia.it) |
| 07:09.27 | *** join/#storm goschtl (n=goschtl@p5B0BBD6F.dip.t-dialin.net) |
| 08:48.51 | *** join/#storm niemeyer (n=niemeyer@nat/canonical/x-4ec69da5a65d5bca) |
| 08:53.11 | *** join/#storm jkakar (n=jkakar@nat/canonical/x-399b04c83634e707) |
| 09:33.13 | *** join/#storm nullguid (n=ilya@87.254.139.66) |
| 09:53.30 | weatherman | storm/default-ordering r237 committed by gustavo@niemeyer.net |
| 09:53.30 | weatherman | Reset sequence on flush, to prevent getting into bigint space. |
| 09:56.36 | *** join/#storm nullie (n=ilya@87.254.139.66) |
| 10:01.15 | weatherman | storm/default-ordering r238 committed by gustavo@niemeyer.net |
| 10:01.15 | weatherman | Add some code comments about the advisory ordering. |
| 10:02.09 | *** join/#storm nullguid_ (n=ilya@87.254.139.66) |
| 10:07.29 | weatherman | storm/default-ordering r239 committed by gustavo@niemeyer.net |
| 10:07.29 | weatherman | Fixed broken zope tests due to the new key in obj_infos, as |
| 10:07.30 | weatherman | correctly detected by Jamu. |
| 10:21.29 | *** join/#storm niemeyer (n=niemeyer@nat/canonical/x-8ba34458fa2ee2a4) |
| 10:24.17 | weatherman | storm/trunk r242 committed by gustavo@niemeyer.net |
| 10:24.17 | weatherman | Merge default-ordering branch [r=jkakar,radix] |
| 10:24.21 | weatherman | This introduces advisory flush ordering. Objects will be flushed in the |
| 10:24.25 | weatherman | order they became dirty if possible (IOW, if there are no other rules |
| 10:24.29 | weatherman | enforcing a different order). |
| 10:26.09 | weatherman | storm/trunk r243 committed by gustavo@niemeyer.net |
| 10:26.09 | weatherman | Added note about the advisory ordering to the NEWS file. |
| 10:58.45 | *** join/#storm bac (n=bac@cpe-065-190-187-178.nc.res.rr.com) |
| 11:02.16 | *** join/#storm bac (n=bac@cpe-065-190-187-178.nc.res.rr.com) |
| 12:13.15 | *** join/#storm bigdo1 (n=scmikes@72-197-8-8-arpa.cust.cinci.current.net) |
| 12:35.58 | *** join/#storm jukart (n=jukart@85-124-221-45.static.xdsl-line.inode.at) |
| 14:08.03 | weatherman | storm/need-for-speed-revenge r258 committed by gustavo@niemeyer.net |
| 14:08.03 | weatherman | Oops.. a dummy commented out import went in by mistake, as Thomas spotted. |
| 14:12.35 | jamesh | niemeyer: ^^^ is this something important? |
| 14:13.19 | therve | jamesh: nope |
| 14:19.08 | jamesh | niemeyer: http://paste.ubuntu.com/21628/ <- this is the other patch I'm applying for LP |
| 14:19.41 | niemeyer | jamesh: Not really.. it was just a line commented out that therve spotted |
| 14:19.51 | jamesh | shrinks down some of the most heavily used objects |
| 14:20.54 | niemeyer | jamesh: The patch looks sensible |
| 14:20.55 | jamesh | the branch is up on Launchpad as ~jamesh/storm/slots |
| 14:21.08 | niemeyer | jamesh: Out of curiosity, have you made any measurements about memory usage? |
| 14:23.24 | jamesh | niemeyer: not really. I need to play with valgrind/massif a bit more |
| 14:24.08 | jamesh | I found out that the latest version of valgrind has changed the Postscript heap vs. time graphs to ascii-art |
| 14:25.36 | niemeyer | jamesh: Hmm.. why was this needed: |
| 14:25.38 | niemeyer | -class ClassAlias(FromExpr): |
| 14:25.38 | niemeyer | +class ClassAlias(object): |
| 14:25.57 | jamesh | with cextensions, most Variable subclasses never allocated a dictionary, but had space in their PyObject struct for one and an instance dict |
| 14:26.18 | jamesh | niemeyer: struct layout incompatibility when doing subclassing it |
| 14:27.12 | jamesh | and the one place where there were it'd matter that ClassAlias was a FromExpr did the same thing if the object wasn't an Expr at all |
| 14:27.58 | niemeyer | jamesh: JoinExpr seems to treat it differently |
| 14:28.12 | jamesh | niemeyer: not really |
| 14:28.13 | niemeyer | jamesh: Or maybe I'm overlooking it.. hmmm |
| 14:28.31 | niemeyer | jamesh: Oh, you're right |
| 14:28.48 | jamesh | while it no longer satisfies "isinstance(arg2, (FromExpr, Alias))", it now satisfies "not isinstance(arg2, Expr)" |
| 14:29.16 | niemeyer | jamesh: Absolutely, thanks for pointing it out |
| 14:30.02 | jamesh | niemeyer: so, adding __slots__=() to ClassAlias got rid of the struct layout incompatibility error, but results in a deallocator incompatibility error |
| 14:30.14 | jamesh | (that was when it was a FromExpr subclass) |
| 14:30.40 | jamesh | seemed easiest to keep the __slots__ optimisations away from object classes |
| 14:32.19 | niemeyer | jamesh: Sounds reasonable indeed |
| 14:36.39 | jamesh | niemeyer: thumper found this bug too: https://bugs.edge.launchpad.net/storm/+bug/241530 |
| 14:36.41 | ubot3 | Malone bug 241530 in storm "Storm does not use references to order flushes of deletes" [Medium,New] |
| 14:37.38 | jamesh | it is something we can work around, so isn't a huge deal but seems like something that should be handled |
| 14:38.04 | jamesh | the "flush objects in order dirtied" rule will probably hide this for the most part though. |
| 14:38.31 | niemeyer | jamesh: I was thinking about this one early today.. I'd like to talk to you at some point about the actual problem seen |
| 14:39.12 | niemeyer | jamesh: I mean, I understand how it affects people in general |
| 14:39.35 | niemeyer | jamesh: I was just considering the various ways in which the objects may be in memory or not, and how that would affect removal |
| 14:40.01 | jamesh | niemeyer: the specific case was "delete a branch and its subscriptions" |
| 14:40.09 | niemeyer | jamesh: I guess this specific bug report is a bit simpler, though.. in the sense that the two objects are actually removed explicitly |
| 14:40.57 | niemeyer | jamesh: So the bug doesn't cascade to its subscriptions in the database? |
| 14:41.27 | jamesh | niemeyer: we're don't do cascaded deletes |
| 14:41.35 | niemeyer | jamesh: Aha, I see |
| 14:42.02 | jamesh | although I sometimes wonder how many rows you'd need to remove in Launchpad with cascade enabled to take out the entire database ... |
| 14:42.42 | niemeyer | jamesh: Hehe :) |
| 14:43.24 | niemeyer | jamesh: I guess the n-to-n tables are always a boundary though |
| 14:44.44 | jamesh | so, in the absence of cascaded deletes, I think you'd need to have the object holding the foreign key reference dirty too |
| 14:44.53 | jamesh | (either deleting the object, or modifying it) |
| 14:46.14 | niemeyer | jamesh: Yeah, we'd basically have to make Storm understand cascading by itself |
| 14:47.27 | jamesh | niemeyer: as I said earlier, the change to not reorder objects in a flush probably solves the immediate problems |
| 14:47.56 | jamesh | thumper was seeing it as an intermittent IntegrityError |
| 14:49.16 | niemeyer | jamesh: That's good to hear.. should give us some time to think about that other feature |
| 17:11.17 | *** join/#storm bigdo1 (n=scmikes@72-197-8-8-arpa.cust.cinci.current.net) |
| 23:59.47 | *** join/#storm jml_ (n=jml@r125-63-186-59.cpe.unwired.net.au) |