Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | expand scope of "no alternate exchange" optimisationbug24504 | Matthias Radestock | 2011-10-21 | 1 | -3/+2 |
| | | | | to cover the "routed somewhere" case too | ||||
* | Revert 910c670238ad (Produce guids a bit more efficiently) | Simon MacMullen | 2011-10-19 | 1 | -7/+7 |
| | |||||
* | 3) Cache results of rabbit_registry:lookup_module() for exchange types in ↵ | Simon MacMullen | 2011-10-19 | 1 | -2/+6 |
| | | | | the process dictionary. | ||||
* | 2) Produce guids a bit more efficiently. | Simon MacMullen | 2011-10-19 | 1 | -7/+7 |
| | | | | (At the cost of limiting us to 2^64 messages published per channel.) | ||||
* | 1) Don't even consider alternate exchanges for exchanges with no args. | Simon MacMullen | 2011-10-19 | 1 | -0/+3 |
| | |||||
* | Merge bug17162. | Simon MacMullen | 2011-10-19 | 3 | -121/+109 |
|\ | |||||
| * | remove parse_properties (again) and rename testbug17162 | Alexandru Scvortov | 2011-10-18 | 2 | -116/+59 |
| | | |||||
| * | small routing optimisation and re-add parse_properties | Alexandru Scvortov | 2011-10-18 | 2 | -1/+58 |
| | | | | | | | | Parse_properties is now only used to test encode_properties. | ||||
| * | cosmetic | Alexandru Scvortov | 2011-10-18 | 1 | -7/+12 |
| | | | | | | | | And fix the '_':2 thing. | ||||
| * | unroll property parser | Alexandru Scvortov | 2011-10-17 | 2 | -59/+46 |
| | | | | | | | | The codegen code is quite ugly now; I'll see what I can do about that. | ||||
* | | yet another attempt to reduce timeouts in CI | Matthias Radestock | 2011-10-19 | 1 | -6/+6 |
| | | | | | | | | | | - catch surplus eagerly - wait longer for confirms, wait shorter for any surplus | ||||
* | | As noted in bug17162 comment 13, this delivers a few percent improvement on ↵ | Simon MacMullen | 2011-10-18 | 1 | -0/+5 |
| | | | | | | | | MM -y 0. | ||||
* | | Merged bug24460 into default | Emile Joubert | 2011-10-18 | 1 | -3/+2 |
|\ \ | |||||
| * | | remove redundant usortbug24460 | Alexandru Scvortov | 2011-10-17 | 1 | -2/+1 |
| | | | | | | | | | | | | It used to be there because it tended to speed things up for some reason. | ||||
| * | | properly sort the queue of acks | Alexandru Scvortov | 2011-10-17 | 1 | -3/+3 |
| | | | |||||
* | | | Merge bug24483 | Simon MacMullen | 2011-10-18 | 8 | -812/+391 |
|\ \ \ | |||||
| * | | | looks like a 200ms wait is still not enough to keep CI happy | Matthias Radestock | 2011-10-17 | 1 | -1/+1 |
| | |/ | |/| | |||||
| * | | merge bug24455 into default | Matthias Radestock | 2011-10-17 | 4 | -786/+365 |
| |\ \ | | |/ | |/| | |||||
| | * | Minor tweaksbug24455 | Matthew Sackman | 2011-10-17 | 1 | -14/+5 |
| | | | |||||
| | * | Having thought about and discussed the code and desires, change things ↵ | Matthew Sackman | 2011-10-17 | 1 | -47/+26 |
| | | | | | | | | | | | | around a bit. | ||||
| | * | cosmetic + correct specs for lqueue:peek{_r} | Matthias Radestock | 2011-10-15 | 1 | -16/+16 |
| | | | |||||
| | * | oops | Matthias Radestock | 2011-10-15 | 1 | -7/+3 |
| | | | |||||
| | * | Revert lqueue so it does not cache ends of queue as we no longer need this ↵ | Matthew Sackman | 2011-10-15 | 1 | -71/+33 |
| | | | | | | | | | | | | to be O(1). | ||||
| | * | refactor: extract delta assertions into one place | Matthias Radestock | 2011-10-15 | 1 | -14/+15 |
| | | | |||||
| | * | doc tidying | Matthias Radestock | 2011-10-15 | 1 | -10/+3 |
| | | | |||||
| | * | merge default into bug24455 | Matthias Radestock | 2011-10-15 | 4 | -772/+429 |
| | |\ | |/ / | |||||
| | * | Because I've got rid of ram_index_count (i.e. betas count), and because the ↵ | Matthew Sackman | 2011-10-12 | 1 | -48/+20 |
| | | | | | | | | | | | | guard around whether or not betas_to_deltas is now based on q2 and q3 len rather than ram_index_count, we no longer need to have the separate gammas_to_deltas fun, which means we should no longer have the problem with it being called too often as we should be able to rely on the chunk stuff working out. | ||||
| | * | gb_sets:from_list(X) is slow. gb_sets:from_ordset(lists:usort(X)) is faster. ↵ | Matthew Sackman | 2011-10-12 | 1 | -8/+9 |
| | | | | | | | | | | | | Just keeping it as a gb_set throughout is fastest still. | ||||
| | * | Update docs | Matthew Sackman | 2011-10-12 | 1 | -29/+70 |
| | | | |||||
| | * | Undo that last change, but still ensure q3 ends on a segment boundary by ↵ | Matthew Sackman | 2011-10-12 | 1 | -16/+9 |
| | | | | | | | | | | | | giving q2 priority over q3 beyond the segment boundary of q3 | ||||
| | * | cosmetic, and be somewhat smarter about deciding the minimum length of q3 ↵ | Matthew Sackman | 2011-10-12 | 2 | -8/+15 |
| | | | | | | | | | | | | (now must end on a segment boundary - minor optimisation) | ||||
| | * | Cache both ends of the queue to give O(1) peek and peek_r | Matthew Sackman | 2011-10-12 | 1 | -37/+68 |
| | | | |||||
| | * | Do not use q1+q4 length when calculating permitted ?s - if you do, then as ↵ | Matthew Sackman | 2011-10-11 | 1 | -5/+4 |
| | | | | | | | | | | | | the queue is slowly drained, even whilst target_ram_count may be growing, you can end up writing more out to disk. Instead, use the target_ram_count directly as an indication of ?s. This is preferable as it indicates the permitted ?s which is obviously suitable for use when calculating the permitted ?s. | ||||
| | * | Add peek and peek_r to lqueue, and use them in needs_timeout. Whilst this is ↵ | Matthew Sackman | 2011-10-11 | 2 | -10/+18 |
| | | | | | | | | | | | | better, it's still O(N) per msg, so we shouldn't actually be doing even this... | ||||
| | * | When under memory pressure, we want to limit the size of q2 and q3, not just ↵ | Matthew Sackman | 2011-10-11 | 2 | -161/+132 |
| | | | | | | | | | | | | of ?s. Simplify calculation of permitted length. Simplify (and correct) expansion of ?. We can now completely drop ram_index_count as we never care about just the number of ?s. | ||||
| | * | Merging default to bug24455 | Matthew Sackman | 2011-10-11 | 6 | -219/+211 |
| | |\ | |||||
| | * | | Ensure that we push ?-bordering ?s into ? eagerly. | Matthew Sackman | 2011-10-11 | 2 | -10/+22 |
| | | | | |||||
| | * | | Merging default into bug24455 | Matthew Sackman | 2011-10-04 | 4 | -648/+271 |
| | |\ \ | |||||
| | | * | | simplify delta calculation in push_betas_to_deltas | Matthias Radestock | 2011-10-03 | 1 | -65/+47 |
| | | | | | |||||
| | | * | | fix indentation | Matthias Radestock | 2011-10-02 | 1 | -2/+1 |
| | | | | | |||||
| | | * | | save a line | Matthias Radestock | 2011-10-02 | 1 | -3/+2 |
| | | | | | |||||
| | | * | | merge heads | Matthias Radestock | 2011-10-02 | 1 | -1/+1 |
| | | |\ \ | |||||
| | | | * | | Make lqueue opaque | Matthew Sackman | 2011-10-02 | 1 | -1/+1 |
| | | | | | | |||||
| | | * | | | refactor: get rid of #merge_funs | Matthias Radestock | 2011-10-02 | 1 | -54/+29 |
| | | | | | | |||||
| | | * | | | some small cosmetic changes and tiny tweaks | Matthias Radestock | 2011-10-02 | 1 | -13/+13 |
| | | |/ / | |||||
| | | * | | Drop unneeded queue impls, tidy lqueue, add specs etc | Matthew Sackman | 2011-10-02 | 6 | -820/+28 |
| | | | | | |||||
| | | * | | Avoid lots of combine_delta calls | Matthew Sackman | 2011-10-02 | 2 | -45/+87 |
| | | | | | |||||
| | | * | | Merging default into bug24455 | Matthew Sackman | 2011-10-02 | 7 | -80/+87 |
| | | |\ \ | |||||
| | | * | | | Avoid delta's low end extending beyond the low end of q3 | Matthew Sackman | 2011-09-30 | 1 | -16/+21 |
| | | | | | | |||||
| | | * | | | Merging default into bug24455 | Matthew Sackman | 2011-09-30 | 5 | -245/+392 |
| | | |\ \ \ |