ARTEMIS-5305 Improving performance of paging with multiple producers #5498
Conversation
clebertsuconic
force-pushed
the
page
branch
10 times, most recently
from
3ce16c6 to
fbf0910
Compare
...rc/main/java/org/apache/activemq/artemis/core/paging/cursor/impl/PageCursorProviderImpl.java
Show resolved
Hide resolved
clebertsuconic
force-pushed
the
page
branch
9 times, most recently
from
d8e2767 to
483f513
Compare
|
Seems to be a heap dump attached to the second commit which is probably not meant to be there |
clebertsuconic
force-pushed
the
page
branch
2 times, most recently
from
afc9278 to
1327ade
Compare
clebertsuconic
force-pushed
the
page
branch
4 times, most recently
from
7388bde to
1608561
Compare
|
I have done a lot of review myself, trying to pretend I was someone else (like Robbie) and review my own stuff.. and I reach a limit ... I can't find anything else :) Tests are extremely stable.. done a lot of testing... |
clebertsuconic
force-pushed
the
page
branch
2 times, most recently
from
574dc22 to
b01f91a
Compare
artemis-server/src/main/java/org/apache/activemq/artemis/core/paging/impl/PagingStoreImpl.java
Outdated
Show resolved
Hide resolved
artemis-server/src/main/java/org/apache/activemq/artemis/core/paging/impl/PagingStoreImpl.java
Outdated
Show resolved
Hide resolved
clebertsuconic
force-pushed
the
page
branch
2 times, most recently
from
0752f44 to
0cdd5fb
Compare
|
Right now there are two operation ways.. one for unit tests (PagingStoreImpl). I will make changes on PagingStoreImpl to account for a SCheduledExecutor. |
|
that way there's only going to be one way to lock this... by being single threaded through the PagingStore::executor (the PageTimedWriter also uses the same executor). |
clebertsuconic
force-pushed
the
page
branch
15 times, most recently
from
e2f4d4f to
88ca17b
Compare
|
I reworked the tests in a way to not require two types of locking. Basically it's pretty much single threaded now (through pageExecutor). |
This is improving how locking works by using a single threaded executor on paging and depaging as well as batching writes instead of directly writing on the files from the producers.
|
@gemmellr can we merge this? I wanted to merge this before the javadoc huge change by Justin... besides I will have to bring this to an internal branch of mine based on a previous version and I need to start working on that.... |
This is improving how locking works by using a single threaded executor on paging and depaging
as well as batching writes instead of directly writing on the files from the producers.