Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | Increase the amount of symmetry.bug25152 | Simon MacMullen | 2012-09-17 | 1 | -7/+7 |
| | |||||
* | bring queue/exchange removal cost back down to O(binding_count) | Matthias Radestock | 2012-09-15 | 1 | -20/+30 |
| | | | | by performing all mnesia read operations before writes | ||||
* | guard deletes on disk tables in order to eliminate superfluos fsyncs | Matthias Radestock | 2012-09-15 | 3 | -5/+23 |
| | | | | | Unfortunately this makes queue deletion O(binding_count^2), so further work is needed. | ||||
* | Merge bug24914 | Simon MacMullen | 2012-09-14 | 5 | -105/+127 |
|\ | |||||
| * | merge bug25118 into default | Tim Watson | 2012-09-14 | 5 | -105/+127 |
| |\ | |||||
| | * | merge default into bug25118bug25118 | Tim Watson | 2012-09-14 | 5 | -105/+127 |
| | |\ | |/ / | |||||
| | * | updated the essay to explain how the sync status works now | Francesco Mazzoli | 2012-09-13 | 1 | -19/+25 |
| | | | |||||
| | * | two little fixes... | Francesco Mazzoli | 2012-09-06 | 1 | -2/+2 |
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | It turns out that the problems I had with the kill-multi test on default was because the messages were start publishing right after the queue was created. I thought that once queue.declare returned, it meant that the queue was present on all nodes, but it wasn't, and for this reason we had the mismatching messages and the purging. Once I discovered that, I went back two my code and immediately discovered two very silly mistakes. I feel silly as well now. I think that bug 25130 still has a reason to extist, but I wouldn't worry about it now. | ||||
| | * | cosmetics | Francesco Mazzoli | 2012-09-04 | 1 | -1/+1 |
| | | | |||||
| | * | cosmetics | Francesco Mazzoli | 2012-09-04 | 2 | -4/+1 |
| | | | |||||
| | * | cosmetics | Francesco Mazzoli | 2012-09-04 | 1 | -1/+1 |
| | | | |||||
| | * | merge default | Francesco Mazzoli | 2012-09-04 | 7 | -110/+222 |
| | |\ | |||||
| | * | | remove differences with default | Francesco Mazzoli | 2012-09-04 | 1 | -10/+12 |
| | | | | |||||
| | * | | track the delta of the depths, and replace `pending_ack' with `depth' in BQ | Francesco Mazzoli | 2012-09-04 | 4 | -71/+52 |
| | | | | | | | | | | | | | | | | The kill-multi test is still failing... | ||||
| | * | | was getting `set_length' instead of `drop' | Francesco Mazzoli | 2012-09-04 | 1 | -1/+1 |
| | | | | |||||
| | * | | fix depth delta in the `publish' instruction | Francesco Mazzoli | 2012-09-04 | 1 | -1/+1 |
| | | | | |||||
| | * | | forgot debug line in | Francesco Mazzoli | 2012-09-03 | 1 | -1/+0 |
| | | | | |||||
| | * | | store the depth of master and slave instead of the unknown pending msgs | Francesco Mazzoli | 2012-09-03 | 2 | -69/+92 |
| | | | | |||||
| | * | | refactoring: make the var match the field | Matthias Radestock | 2012-08-31 | 1 | -8/+7 |
| | | | | |||||
| | * | | refactoring: inline helper function | Matthias Radestock | 2012-08-31 | 1 | -24/+25 |
| | | | | |||||
| | * | | cosmetic | Matthias Radestock | 2012-08-31 | 1 | -3/+2 |
| | | | | |||||
| | * | | style | Francesco Mazzoli | 2012-08-31 | 1 | -5/+3 |
| | | | | |||||
| | * | | revert `set_synchronized' => `update_unknown_pending' | Francesco Mazzoli | 2012-08-31 | 1 | -22/+22 |
| | | | | | | | | | | | | | | | | | | | | Matthias likes the old names better, arguing that the main purpose of that function is to do that mnesia transaction. | ||||
| | * | | `set_synchronized' => `update_unknown_pending' | Francesco Mazzoli | 2012-08-31 | 1 | -22/+22 |
| | | | | |||||
| | * | | `set_length' => `drop' | Francesco Mazzoli | 2012-08-31 | 1 | -2/+2 |
| | | | | |||||
| | * | | `external_pending' => `unknown_pending' | Francesco Mazzoli | 2012-08-31 | 1 | -6/+9 |
| | | | | |||||
| | * | | use `set_synchronized' instead than manually increasing | Francesco Mazzoli | 2012-08-31 | 1 | -6/+2 |
| | | | | |||||
| | * | | simplify `set_length' case, fixing the ToDrop < 0 branch | Francesco Mazzoli | 2012-08-31 | 1 | -20/+16 |
| | | | | | | | | | | | | | | | | The unknown pending counter wasn't updated correctly | ||||
| | * | | add the delta in `set_synchronized', fixes the call in `set_length' | Francesco Mazzoli | 2012-08-31 | 1 | -3/+3 |
| | | | | |||||
| | * | | `set_synchronized' accepts the difference instead of separate args | Francesco Mazzoli | 2012-08-31 | 1 | -7/+7 |
| | | | | |||||
| | * | | do not track external pendings until we receive `length' | Francesco Mazzoli | 2012-08-31 | 1 | -2/+7 |
| | | | | | | | | | | | | | | | | Otherwise, we might break some assertions. | ||||
| | * | | actually, it looks like the slave might receive messages before `length' | Francesco Mazzoli | 2012-08-31 | 1 | -1/+1 |
| | | | | | | | | | | | | | | | | | | | | This is not a problem, since the worse that can happen is that we have a bogus sync state in the until `length' is received. | ||||
| | * | | `length' should be the first instructions, lets assert that | Francesco Mazzoli | 2012-08-31 | 1 | -1/+1 |
| | | | | |||||
| | * | | get the external pending acks at the beginning | Francesco Mazzoli | 2012-08-31 | 2 | -5/+7 |
| | | | | |||||
| | * | | add a callback to backing queue to get the number of pending acks | Francesco Mazzoli | 2012-08-31 | 3 | -7/+17 |
| | | | | |||||
| | * | | matthias doesn't like the nested `set_synchronised' :( | Francesco Mazzoli | 2012-08-30 | 1 | -20/+18 |
| | | | | |||||
| | * | | take into account requeues when setting synch state for slaves | Francesco Mazzoli | 2012-08-30 | 2 | -58/+65 |
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | To do this, keep count all the fetches we've seen that require an ack for messages we don't have (e.g. when the queue we have is shorter than on the master). We then decrease this counter appropriately when requeueing, acking, and set_length'ing. Given this, we can deem the slave synced only when the length is the same *and* the counter described above is 9 - there are no pending acks on the master for messages we don't have. I might have missed something (I barely tested this) but it seems to do the trick. | ||||
* | | | | fix comment in node_monitor, keep {error, enoent} in `try_read_file'bug24914 | Francesco Mazzoli | 2012-09-14 | 1 | -12/+8 |
| | | | | |||||
* | | | | merge default | Francesco Mazzoli | 2012-09-14 | 9 | -822/+1299 |
|\ \ \ \ | |/ / / |/| | | | |||||
| * | | | get cluster nodes from mnesia when legacy/non existant status file | Francesco Mazzoli | 2012-09-13 | 1 | -2/+8 |
| | | | | | | | | | | | | | | | | | | | | This is necessary because the upgrade process needs to remove the replicas of the schemas. | ||||
| * | | | check the cluster consistency after the upgrade | Francesco Mazzoli | 2012-09-13 | 3 | -10/+12 |
| | | | | |||||
| * | | | always call init_db with all nodes | Francesco Mazzoli | 2012-09-12 | 1 | -6/+6 |
| | | | | |||||
| * | | | more informative and useful error messages | Francesco Mazzoli | 2012-09-12 | 2 | -4/+10 |
| | | | | |||||
| * | | | do not give up when finding an inconsistent node | Francesco Mazzoli | 2012-09-11 | 1 | -7/+10 |
| | | | | |||||
| * | | | oops | Francesco Mazzoli | 2012-09-11 | 1 | -2/+2 |
| | | | | |||||
| * | | | split cluster status file and running nodes file | Francesco Mazzoli | 2012-09-11 | 2 | -118/+97 |
| | | | | |||||
| * | | | Oops | Simon MacMullen | 2012-09-10 | 1 | -1/+1 |
| | | | | |||||
| * | | | Attempt at simplifying check_cluster_consistency/0, but I am having trouble ↵ | Simon MacMullen | 2012-09-10 | 1 | -36/+40 |
| | | | | | | | | | | | | | | | | with systest. | ||||
| * | | | Typos and rephrasing | Simon MacMullen | 2012-09-07 | 3 | -6/+6 |
| | | | | |||||
| * | | | typo in rabbitmqctl man page | Francesco Mazzoli | 2012-09-05 | 1 | -1/+1 |
| | | | |