diff options
| author | Matthew Sackman <matthew@lshift.net> | 2009-11-09 13:46:19 +0000 |
|---|---|---|
| committer | Matthew Sackman <matthew@lshift.net> | 2009-11-09 13:46:19 +0000 |
| commit | 5fd1c1493c488631c56d180117c055c2f23ddb7c (patch) | |
| tree | 89d70e1d4208cfa9ea4c9f7d81c1faf27b6614f6 /Makefile | |
| parent | e3598c965896854b5f23d1fa42abbb78273fee15 (diff) | |
| download | rabbitmq-server-git-5fd1c1493c488631c56d180117c055c2f23ddb7c.tar.gz | |
Made the tests work again. Also one tiny cosmetic in vq. However, uncovered major mistake in VQ which is that currently, when the prefetcher starts, that may empty q3. A subsequent publish may then go straight to Q4, thus overtaking the previous msgs. Even worse, when the prefetcher is drained, there is no attempt to join it into the existing q4, it just replaces it. What should happen is that the existence of the prefetcher is treated as if both q3 and q4 are non empty. This makes sense, because there are some ways in which the prefetcher can exit, returning entries for both q4 and q3. Thus pubs that happen after the prefetcher is started must go to q1/q2/γ, and so we know that entries already in q4 when the prefetcher is drained must have got there before the prefetcher was started. Finally, when the prefetcher is drained, if γ is empty, q2 and q3 can be joined, and if q2, γ and q3 are empty, q1 and q4 can be joined. Or something like that.
Diffstat (limited to 'Makefile')
0 files changed, 0 insertions, 0 deletions
