Thu Jun 14 19:00:37 PDT 2007
- Previous message: [Slony1-general] Readying 1.2.10
- Next message: [Slony1-general] Testing going well
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
On 6/14/2007 11:03 AM, Jan Wieck wrote: > On 6/13/2007 7:55 PM, Steven Singer wrote: >> On Wed, 13 Jun 2007, Christopher Browne wrote: >> >> >> I've downloaded the tarball can compiled against a 8.3 snapshot on Linux and >> I'm still getting periodic failures of the testddl unit test. >> >> Usually its a <stdin>:9: timeout exceeded while waiting for event confirmation > > Not sure where this one is coming from. There seems to be only one case > of using WAIT FOR EVENT in that test and that does not have a timeout (I > must obviously be missing something). Yepp, I was missing something. WAIT FOR EVENT has a timeout default of 600 seconds. Which of course means that if node 3 never starts to subscribe because of having received the ENABLE_SUBSCRIPTION from node 2 instead of node 1 (which depends on a race condition), then this is exactly what is going to happen. I have fixed the bug in the copy_set() logic but still need to fix the rebuildListenEntries() madness. Jan > >> >> I keep seeing messages along the line in the slon.2.log like: >> 2007-06-13 19:25:20 EDT WARN remoteWorkerThread_1: copy set: data provider >> 1 only on sync -1 - sleep 5 seconds >> >> Is this the problem Jan is working on or is it something else? >> Has anyone else tried to run the testddl unit test a handful of times with >> better luck? > > I wasn't up to now ;-) > > The problem here seems to be that the sl_listen is entirely messed up > which as a side effect triggers what appears to be a bug. > > Why rebuildListenEntries() generates 10 entries for a 3 node setup for > sure needs some investigation. The real underlying problem is that slon > at copy_set() time looks from where it got the event and if that node is > not the one providing the data, it checks if it has seen a confirmation > of the ENABLE_SUBSCRIPTION from the data provider. The thing here is > that data provider is actually the event origin and nodes don't confirm > their own events. So the check needs to exclude the case where the event > origin is the data provider. > > > Jan > -- #======================================================================# # It's easier to get forgiveness for being wrong than for being right. # # Let's break this rule - forgive me. # #================================================== JanWieck at Yahoo.com #
- Previous message: [Slony1-general] Readying 1.2.10
- Next message: [Slony1-general] Testing going well
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Slony1-general mailing list