summaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAgeFilesLines
* Keep the two handle_exception()s together.Simon MacMullen2013-12-191-3/+3
|
* Treat connection errors on write the same as on read.Simon MacMullen2013-12-172-1/+4
|
* Merged bug25925 into stableEmile Joubert2013-12-160-0/+0
|\
* | Strip extra parameter when installing serviceEmile Joubert2013-12-131-2/+10
|/
* Added tag rabbitmq_v3_2_2 for changeset d3f84b20c376Emile Joubert2013-12-110-0/+0
|
* Add changelog entries for 3.2.2rabbitmq_v3_2_2Emile Joubert2013-12-112-0/+9
|
* Merge bug25899Simon MacMullen2013-12-061-2/+12
|\
| * Merge bug25912Simon MacMullen2013-12-060-0/+0
| |\
| * | ExplainSimon MacMullen2013-12-061-0/+2
| | |
| * | Make sure that our last event before going sorta-idle is not one in which we ↵Simon MacMullen2013-12-051-2/+10
| |/ | | | | | | claim to be in flow control.
* | Remove unnecessary test.Matthew Sackman2013-12-061-12/+1
| |
* | Undo some changes.Matthew Sackman2013-12-062-42/+11
| |
* | Rather than just querying a single remote node, query them all, and ↵Matthew Sackman2013-12-042-11/+42
|/ | | | | | accomodate remote nodes that might be running older versions than ourselves, provided we can find someone who is running the same as us. I did try just limiting Nodes to Nodes with the rabbit process running on them. That seemed to break the tests, and trying to figure out why by looking at rabbit_mnesia made me want to slit my wrists and other body parts. Thus I wrote more code instead.
* Merge bug25901Simon MacMullen2013-12-030-0/+0
|\
* | Ooops.Matthew Sackman2013-11-281-2/+2
| |
* | Introduce halt_on_upgrade_failure app env param and set it to true. Only ↵Matthew Sackman2013-11-282-2/+8
|/ | | | throw an error instead of halting if it has been set to false.
* Merge bug25897Simon MacMullen2013-11-280-0/+0
|\
* | Ensure that we always interleave start / stop / start / stop.Simon MacMullen2013-11-281-1/+5
| |
* | Do nothing if the rabbit application is already running. Similarly, don't ↵Simon MacMullen2013-11-281-13/+19
|/ | | | explode if the boot marker is there, just do nothing.
* Merge bug25873Simon MacMullen2013-11-220-0/+0
|\
* \ Merge in stableSimon MacMullen2013-11-223-40/+65
|\ \ | |/ |/|
| * *cough*Simon MacMullen2013-11-121-1/+1
| |
| * Restore synchronous addition of slaves in the queue.declare case.Simon MacMullen2013-11-124-26/+40
| |
| * Forgot to add a specSimon MacMullen2013-11-111-0/+1
| |
| * Re-add the log invocations we lost. This does lead to the stale master pid ↵Simon MacMullen2013-11-112-3/+7
| | | | | | | | warning getting logged on a different node (the slave node rather than the node that attempted to start the slave) but I doubt we care about that.
| * Port the deadlock-breaking code from bug 25852.Simon MacMullen2013-11-112-30/+40
| |
* | Merge bug25867Simon MacMullen2013-11-223-37/+51
|\ \
| * | Oops, that should have gone in bug 25415Simon MacMullen2013-11-141-21/+0
| | |
| * | Merged bug25870 into stableEmile Joubert2013-11-110-0/+0
| |\ \ | | |/ | |/|
| * | Remove 'evil' receive block.Simon MacMullen2013-11-111-14/+18
| | |
| * | Cope with clients dying.Simon MacMullen2013-11-082-16/+47
| |/
| * Added tag rabbitmq_v3_2_1 for changeset fde7d8bb464aSimon MacMullen2013-11-070-0/+0
| |
* | More explanationSimon MacMullen2013-11-221-0/+2
| |
* | Improve comment.Simon MacMullen2013-11-181-3/+4
| |
* | ExplainSimon MacMullen2013-11-061-0/+3
| |
* | Maybe sync on slave promotion.Simon MacMullen2013-11-061-3/+4
|/
* Merged bug25849 into stablerabbitmq_v3_2_1Emile Joubert2013-11-011-0/+5
|\
| * sp.Emile Joubert2013-11-011-1/+1
| |
| * We can get down_from_gm twice in a row, don't blow up if we do.Simon MacMullen2013-10-281-0/+5
| |
* | Merged bug25618 into stableEmile Joubert2013-11-011-16/+25
|\ \
| * | Remove use of gm:leave/1: it does almost nothing, master does not invoke it, ↵Simon MacMullen2013-10-291-7/+2
| | | | | | | | | | | | and it's racy anyway.
| * | Attempt to clean up actions when terminating a slave. Slave termination now ↵Simon MacMullen2013-10-291-15/+29
| | | | | | | | | | | | never deletes a queue.
* | | That async_dirty is not needed, there's one in rabbit_runtime_parameters:list/2.Simon MacMullen2013-11-011-5/+1
| | |
* | | Merged bug25846 into stableEmile Joubert2013-10-3111-22/+54
|\ \ \ | |/ / |/| |
| * | Revivify stableSimon MacMullen2013-10-311-1/+1
| | |
| * | Merged bug25842 into stableEmile Joubert2013-10-311-2/+2
| |\ \
| | * | Non-Windows 32 bit platforms tend to only give you 2GB address space per ↵Simon MacMullen2013-10-301-2/+2
| |/ / | | | | | | | | | process.
| * | Merge bug25845Simon MacMullen2013-10-291-1/+1
| |\ \
| | * | Use 580 seconds as default heartbeat intervalMichael Klishin2013-10-281-1/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | 240 seconds gives a theoretiacl 2.5 decrease in server load (over the previous default of 600 seconds), while 580 gives a ~3% increase. In a test that involved watching CPU usage 1000 idle connections I observe effectively no difference between 600 and 580 values. 580 seconds is enough to cause network activity every 290 (< 300) seconds and hopefully make F5 gear with default timeout settings happy.
| | * | Lower default heartbeat setting to 240 secondsMichael Klishin2013-10-271-2/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | F5 load balancers by default will close TCP connections that have been idle for 300 seconds. The choice of the new 240 seconds interval is not particularly scientific: any < 10 minutes interval would work, given that heartbeats are sent at roughly 1/2 the interval. As far as performance overhead from heartbeats goes, 240 is about as good as 590.
View Lorry Controller Status