StreamQueue: only unmap unused segments #557
Conversation
|
Don't like the 60s sleep.. I think we can do better than that. Maybe a channel that consumers can notify when they switch segment. |
A channel that consumers notify when switching segments sounds good to me. Should we unmap on consumer disconnect as well? |
|
Ah, when consumer is disconnecting too, good catch, we don't do that in the normal queues either currently. |
|
Main benchmark PR benchmark Keep in mind, these numbers are not representative of LavinMQ's peak performance. |
carlhoerberg
force-pushed
the
stream-queue-unmap
branch
from
fb44bff to
5b1b092
Compare
Previous behavior could unmap segments that other consumers were still delivering from, causing segfaults.
Digest would otherwise with its small IO.copy buffer use the double buffering.
FileUtil.rm_rf would otherwise possibly have to iterate over a lot of files
carlhoerberg
force-pushed
the
stream-queue-unmap
branch
from
2b1534d to
f1a9d7f
Compare
Memory should never be allocated in a finalize method so can't have the risk of raising something there
File.write in 1.9.2 allocates a write buffer, which is unessecary
carlhoerberg
force-pushed
the
stream-queue-unmap
branch
from
9f94f1a to
5d845eb
Compare
Previous behavior could unmap segments that other consumers were still delivering from, causing segfaults.