summaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAgeFilesLines
* Merge defaultSimon MacMullen2012-10-0210-157/+326
|\
| * Don't throw, rabbit_amqqueue:with/2 will eat it.Simon MacMullen2012-09-281-1/+1
| |
| * Ignore if the node is down.Simon MacMullen2012-09-281-1/+9
| |
| * Assert a bit here too.Simon MacMullen2012-09-281-1/+1
| |
| * We want to catch here, if sup:start_child exits the rabbit_amqqueue:with we ↵Simon MacMullen2012-09-281-1/+1
| | | | | | | | are in will convert any exit to not_found. Which is unlikely to be helpful
| * Merge in defaultSimon MacMullen2012-09-2810-155/+316
| |\
| | * Merge in defaultSimon MacMullen2012-09-2510-156/+316
| | |\
| | | * Handle the case where we go from unmirrored(A) to nodes(B, C).Simon MacMullen2012-09-201-1/+8
| | | |
| | | * "nodes" policy should not suggest nodes which are not running. Also if nodes ↵Simon MacMullen2012-09-202-5/+19
| | | | | | | | | | | | | | | | policy is completely unconnected with reality, just keep the master alive rather than blow up.
| | | * That case was originally written to cover the unmirrored case - but actually ↵Simon MacMullen2012-09-201-2/+0
| | | | | | | | | | | | | | | | the queue might be mirrored but with no slaves, or about to become mirrored, or whatever. And our is_process_alive() check serves the same purpose.
| | | * Start new slaves in response to gm deaths a bit later, to prevent deadlock. ↵Simon MacMullen2012-09-204-32/+27
| | | | | | | | | | | | | | | | Also remove {add,drop}_mirror/3 as not used, and don't export drop_mirror/2 as not used outside the module.
| | | * That hack should not be needed any more.Simon MacMullen2012-09-191-13/+7
| | | |
| | | * Merge in defaultSimon MacMullen2012-09-1915-355/+364
| | | |\
| | | * | Pull our attempts to list all nodes out of transactions and loops.Simon MacMullen2012-09-181-5/+15
| | | | |
| | | * | SimplifySimon MacMullen2012-09-171-5/+1
| | | | |
| | | * | Cope with the fact that rabbit_mnesia:running_clustered_nodes/0 now does not ↵Simon MacMullen2012-09-171-2/+11
| | | | | | | | | | | | | | | | | | | | include node() during boot.
| | | * | Merge defaultSimon MacMullen2012-09-179-134/+271
| | | |\ \
| | | | * \ Merge defaultSimon MacMullen2012-09-149-134/+271
| | | | |\ \
| | | | | * | cosmeticMatthias Radestock2012-09-051-4/+2
| | | | | | |
| | | | | * | Also exclude DeadNodes from the list of new nodes to start mirrors on; we ↵Simon MacMullen2012-09-051-1/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | have checked for running_clustered_nodes() but that could be out of date.
| | | | | * | Cosmetic: give that comment room to breath.Simon MacMullen2012-09-051-20/+21
| | | | | | |
| | | | | * | Call add_mirror/2 before drop_mirror/2, and explain why.Simon MacMullen2012-09-051-1/+11
| | | | | | |
| | | | | * | GahSimon MacMullen2012-09-052-4/+2
| | | | | | |
| | | | | * | No we don't. The point is that if a mirror is dropping out of the queue, ↵Simon MacMullen2012-09-052-2/+0
| | | | | | | | | | | | | | | | | | | | | | | | | | | | does it need to explicitly leave the GM group? The answer is no: the master is linked to the coordinator is linked to the GM / the slave is linked to the GM.
| | | | | * | Add specs, remove misleading "%% temp" comment.Simon MacMullen2012-09-053-9/+15
| | | | | | |
| | | | | * | Reduce difference to defaultSimon MacMullen2012-09-051-3/+3
| | | | | | |
| | | | | * | merge default into bug24908Matthias Radestock2012-09-049-136/+262
| | | | | |\ \
| | | | | | * | more reification, plus cosmeticsMatthias Radestock2012-09-041-18/+13
| | | | | | | |
| | | | | | * | Reify a touch more.Simon MacMullen2012-09-041-5/+6
| | | | | | | |
| | | | | | * | Suggest queue nodes based on running nodes, otherwise (at the very least) we ↵Simon MacMullen2012-08-231-1/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | can immediately try to start a mirror on a node that has just gone down.
| | | | | | * | Fix broken mergeSimon MacMullen2012-08-231-2/+2
| | | | | | | |
| | | | | | * | s/at-least/exactly/gSimon MacMullen2012-08-222-12/+12
| | | | | | | |
| | | | | | * | Merge defaultSimon MacMullen2012-08-2110-173/+169
| | | | | | |\ \
| | | | | | * | | at-least mode can imply that we need to start slaves in response to slaves ↵Simon MacMullen2012-08-171-10/+30
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | dying elsewhere. So do that.
| | | | | | * | | Merge defaultSimon MacMullen2012-08-159-125/+234
| | | | | | |\ \ \
| | | | | | | * | | Reify something that was previously a little bit magic.Simon MacMullen2012-08-072-6/+13
| | | | | | | | | |
| | | | | | | * | | Support changing master (umm, which actually turned out to be a lot easier ↵Simon MacMullen2012-08-071-10/+7
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | than expected, thanks Matthew).
| | | | | | | * | | Take account of the current nodes when selecting new nodes, and some unit ↵Simon MacMullen2012-08-072-29/+70
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | tests for the node selection logic.
| | | | | | | * | | Merge in defaultSimon MacMullen2012-08-078-120/+184
| | | | | | | |\ \ \
| | | | | | | | * | | Various fixes.Simon MacMullen2012-07-092-17/+9
| | | | | | | | | | |
| | | | | | | | * | | Merge bug25048Simon MacMullen2012-07-098-123/+195
| | | | | | | | |\ \ \
| | | | | | | | | * \ \ Merge bug25048Simon MacMullen2012-07-098-123/+195
| | | | | | | | | |\ \ \
| | | | | | | | | | * | | Dynamic change of HA policy. Promotion from non-HA to master, and demotion ↵Simon MacMullen2012-07-065-32/+112
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | from master to non-HA as appropriate. There will still be problems if the policy changes such that the master needs to change - I'm not sure we should even allow this.
| | | | | | | | | | * | | Rename function, remove utter rubbish.Simon MacMullen2012-07-063-15/+15
| | | | | | | | | | | | |
| | | | | | | | | | * | | Rough sketch of dynamic queue HA-ness:Simon MacMullen2012-07-057-92/+89
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | * Remove #amqqueue.mirror_nodes, we will always derive this from policy * Remove everything to do with x-ha-* arguments * Abstract a bit more stuff into rabbit_mirror_queue_misc * Add a new "at-least" mode This works! Sort of. Changing policies on the fly, changing HAness altogether, having "at-least" set up a new mirror when one disappears, and probably some other things do not work.
* | | | | | | | | | | | | cosmeticMatthias Radestock2012-10-011-2/+1
| | | | | | | | | | | | |
* | | | | | | | | | | | | cosmetic(ish)Matthias Radestock2012-10-011-13/+4
| | | | | | | | | | | | |
* | | | | | | | | | | | | remove superfluous commentMatthias Radestock2012-10-011-4/+0
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | We already say that stuff from MA gets requeued. Furthermore, the master doesn't have an 'MA' field as such - the information is kept in the backing queue.
* | | | | | | | | | | | | cosmetic: remove comment that no longer appliesMatthias Radestock2012-10-011-1/+0
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This should have been done when consumer cancellation notifications were introduced.
* | | | | | | | | | | | | stylistic consistencyMatthias Radestock2012-10-011-1/+1
| | | | | | | | | | | | |
View Lorry Controller Status