Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | oopsbug25386 | Matthias Radestock | 2013-01-07 | 1 | -4/+3 |
| | |||||
* | get rid of #tx{} | Matthias Radestock | 2013-01-07 | 1 | -39/+38 |
| | | | | and fix uncommitted_acks info item | ||||
* | simplify & document ordering | Matthias Radestock | 2013-01-07 | 1 | -17/+19 |
| | |||||
* | record pending acks in a queue rather than set in queue process | Matthias Radestock | 2013-01-07 | 2 | -28/+53 |
| | | | | | | | | Just as we do in the channel; this is more efficient for the typical ack-in-order access pattern. We depend on tags being passed to the queue process in order when ack'ing/rejecting. This requires some slightly fiddly code in the channel. | ||||
* | optimise ack collection | Matthias Radestock | 2013-01-06 | 1 | -3/+9 |
| | | | | for the common case of ack'ing/reject'ing the oldest tag | ||||
* | optimise rabbit_channel:ack/2 | Matthias Radestock | 2013-01-06 | 1 | -29/+26 |
| | | | | | | | | - moving the ?INCR_STATS call inside the fold_per_queue fun reduces consing when stats are disabled - since this was the only call to fold_per_queue where we cared about the result, and we no longer do, we can switch the 'fold' to 'foreach' | ||||
* | restrict previous optimisation, for better workingness | Matthias Radestock | 2013-01-06 | 1 | -1/+3 |
| | |||||
* | cosmetic | Matthias Radestock | 2013-01-06 | 1 | -1/+1 |
| | |||||
* | optimise "no confirms" case of rabbit_amqqueue_process:discard | Matthias Radestock | 2013-01-06 | 1 | -4/+7 |
| | |||||
* | optimise "route to no queues" path | Matthias Radestock | 2013-01-06 | 1 | -0/+4 |
| | |||||
* | common-case optimisations for delegate:invoke[_no_result] | Matthias Radestock | 2013-01-06 | 1 | -0/+12 |
| | |||||
* | optimise rabbit_amqqueue:qpids/1 common case | Matthias Radestock | 2013-01-05 | 1 | -0/+2 |
| | |||||
* | optimisation: improve performance of permission cache | Matthias Radestock | 2013-01-05 | 1 | -9/+7 |
| | | | | | | Don't update the cache when the permission is already in it. This saves list munching and a 'put' at the expense of no longer being strictly LRU, but that only affects pathological cases. | ||||
* | refactor gen_server2 debug handling | Matthias Radestock | 2013-01-05 | 1 | -38/+22 |
| | | | | | | - more uniformity - less code duplication - less closure creation -> performance increase | ||||
* | optimisation: shrink gen_server2 state slightly | Matthias Radestock | 2013-01-05 | 1 | -18/+14 |
| | | | | for a (very) modest performance gain, and slightly neater code | ||||
* | refactor queue initialisation | Matthias Radestock | 2013-01-04 | 1 | -34/+16 |
| | | | | | - get rid of all the 'undefined' setting, since that's the default - extract commonality of state initialisation | ||||
* | merge bug24407 into default | Matthias Radestock | 2013-01-04 | 8 | -30/+412 |
|\ | |||||
| * | tweak loggingbug24407 | Matthias Radestock | 2013-01-04 | 1 | -2/+1 |
| | | |||||
| * | cosmetic(ish) | Matthias Radestock | 2013-01-04 | 1 | -1/+1 |
| | | |||||
| * | cosmetic | Matthias Radestock | 2013-01-03 | 1 | -3/+2 |
| | | |||||
| * | simplify syncer | Matthias Radestock | 2013-01-03 | 1 | -33/+23 |
| | | | | | | | | we don't need to track monitors | ||||
| * | make credit waiting less brittle | Matthias Radestock | 2013-01-03 | 2 | -12/+11 |
| | | | | | | | | | | | | | | | | | | | | | | | | We were relying on running out of credit for all slaves *simultaneously*, which requires in-depth knowledge of the credit flow logic and that no other credit-requiring messages are sent to a slave prior to this. Fortunately, since we are running in a fresh separate process we can simply handle *any* credit bumping message and *any* DOWN message. As a bonus we can revert to making the type of the bump msg opaque. | ||||
| * | fix bug found by dialyzer | Matthias Radestock | 2013-01-03 | 1 | -2/+2 |
| | | |||||
| * | bring up to date with 'default' | Matthias Radestock | 2013-01-03 | 1 | -1/+1 |
| | | |||||
| * | merge default into bug24407 | Matthias Radestock | 2013-01-03 | 25 | -615/+677 |
| |\ | |||||
| * | | Specs. | Simon MacMullen | 2013-01-03 | 3 | -1/+22 |
| | | | |||||
| * | | Don't use a closure for the usual cluster upgrade reasons. | Simon MacMullen | 2013-01-03 | 1 | -4/+7 |
| | | | |||||
| * | | API consistency. | Simon MacMullen | 2013-01-03 | 2 | -12/+8 |
| | | | |||||
| * | | Consistency with the real slave. | Simon MacMullen | 2013-01-03 | 1 | -2/+2 |
| | | | |||||
| * | | Explain why | Simon MacMullen | 2013-01-03 | 1 | -0/+1 |
| | | | |||||
| * | | cosmetic(ish) refactors on mq_sync:master_send | Matthias Radestock | 2012-12-30 | 1 | -10/+9 |
| | | | | | | | | | | | | | | | | | | - more sensible arg order (a folding function should generally take the element first and the acc last) - somewhat neater handling of the acc | ||||
| * | | correct docs to reflect idempotence | Matthias Radestock | 2012-12-30 | 1 | -3/+2 |
| | | | |||||
| * | | consistency | Matthias Radestock | 2012-12-30 | 1 | -1/+1 |
| | | | |||||
| * | | It's important to respond to this too. | Simon MacMullen | 2012-12-14 | 1 | -1/+4 |
| | | | |||||
| * | | Make syncing idempotent. | Simon MacMullen | 2012-12-06 | 2 | -3/+2 |
| | | | |||||
| * | | Add rabbitmqctl sync_queue, and make it synchronous. Also fix the fact that ↵ | Simon MacMullen | 2012-12-06 | 6 | -14/+67 |
| | | | | | | | | | | | | we were treating "no queues to sync" as a 'normal' crash of the syncer, which... no. | ||||
| * | | Normalise credit flow quantities. | Simon MacMullen | 2012-12-05 | 1 | -2/+2 |
| | | | |||||
| * | | correct docs for final handshake | Matthias Radestock | 2012-11-29 | 1 | -0/+1 |
| | | | |||||
| * | | handle the case of the Syncer dying right at the end | Matthias Radestock | 2012-11-29 | 1 | -15/+20 |
| | | | | | | | | | | | | | | | | | | which could previously leave the master blocked, waiting for 'next'. And move the unlinking, which allows us to ensure we don't end up with stray 'EXIT's. | ||||
| * | | cosmetic | Matthias Radestock | 2012-11-29 | 1 | -4/+4 |
| | | | |||||
| * | | extract logging | Matthias Radestock | 2012-11-29 | 2 | -26/+21 |
| | | | |||||
| * | | no more unused vars | Matthias Radestock | 2012-11-29 | 1 | -1/+1 |
| | | | |||||
| * | | Don't do anything if we decided not to do anything. | Simon MacMullen | 2012-11-29 | 2 | -1/+2 |
| | | | |||||
| * | | Merge heads | Simon MacMullen | 2012-11-29 | 1 | -26/+17 |
| |\ \ | |||||
| | * | | cosmetic, plus correctly identify provenance | Matthias Radestock | 2012-11-29 | 1 | -26/+17 |
| | | | | |||||
| * | | | Set depth_delta to undefined if something goes wrong. Or indeed if something ↵ | Simon MacMullen | 2012-11-29 | 1 | -4/+3 |
| |/ / | | | | | | | | | | goes right, but in that case we are about to overwrite it. | ||||
| * | | Update diagram. | Simon MacMullen | 2012-11-29 | 1 | -2/+4 |
| | | | |||||
| * | | Make sure newly-started slaves don't respond and confuse things. | Simon MacMullen | 2012-11-29 | 2 | -3/+6 |
| | | | |||||
| * | | Call get_parent/0 less often... | Simon MacMullen | 2012-11-29 | 1 | -6/+6 |
| | | | |||||
| * | | Respond to FHC. | Simon MacMullen | 2012-11-29 | 1 | -1/+7 |
| | | |