Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

common/options: bluefs_buffered_io=true by default #20542

Merged
merged 1 commit into from Mar 5, 2018

Conversation

Projects
None yet
4 participants
@liewegas
Copy link
Member

commented Feb 22, 2018

This sends bluefs (rocksdb) io through the host's page cache. It reduces
some of our exposure to bad cache size tuning (since misses in rocksdb's
cache may be caught by the host's page cache) and also mitigates the
cache invalidation that happens during compression.

I originally pushed to get this disabled with an eye toward removing
support for buffered IOs completely, and to avoid a behavior disparity
with the SPDK backend (which has no page cache). In reality this code
isn't going away, and this has a measurable positive impact for users
today.

See 7f568fa for when it was originally
turned off.

Signed-off-by: Sage Weil sage@redhat.com

common/options: bluefs_buffered_io=true by default
This sends bluefs (rocksdb) io through the host's page cache.  It reduces
some of our exposure to bad cache size tuning (since misses in rocksdb's
cache may be caught by the host's page cache) and also mitigates the
cache invalidation that happens during compression.

I originally pushed to get this disabled with an eye toward removing
support for buffered IOs completely, and to avoid a behavior disparity
with the SPDK backend (which has no page cache).  In reality this code
isn't going away, and this has a measurable positive impact for users
today.

See 7f568fa for when it was originally
turned off.

Signed-off-by: Sage Weil <sage@redhat.com>
@yuriw

This comment has been minimized.

Copy link
Contributor

commented Mar 1, 2018

@markhpc

markhpc approved these changes Mar 5, 2018

Copy link
Member

left a comment

I think in the short/medium term this is going to help more than it hurts. Enabling it has resulted in higher performance in a variety of tests.

@yuriw yuriw merged commit 8eb3c1f into ceph:master Mar 5, 2018

5 checks passed

Docs: build check OK - docs built
Details
Signed-off-by all commits in this PR are signed
Details
Unmodified Submodules submodules for project are unmodified
Details
make check make check succeeded
Details
make check (arm64) make check succeeded
Details
@gregsfortytwo

This comment has been minimized.

Copy link
Member

commented Mar 5, 2018

Did anybody do a review to see if it's even plausible that the buffered IO option is stable? Looks like we haven't tested it in nightlies since August 2016, and I'm noticing some inconsistencies around how that buffered flag is used in the IO functions.

@markhpc

This comment has been minimized.

Copy link
Member

commented Mar 5, 2018

gregsfortytwo: I'm not opposed to reverting this if you have concerns!

@liewegas liewegas deleted the liewegas:wip-bluefs-buffered-io branch Mar 5, 2018

@gregsfortytwo

This comment has been minimized.

Copy link
Member

commented Mar 5, 2018

@liewegas said he'd look into it after lunch, which is about as much as I was hoping for. (I don't know anything specific about the bluestore internals.)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.