MySQLOnRocksDB/mysql-5.6
forked from facebook/mysql-5.6

Loading…
Owner
mdcallag
commented
|
|
maykov |
Making shutdowns faster by calling DisownData
…
Summary: As above , see the issue #59 Test Plan: I tested for correcntess by looking at the log file and didn't see anything suspicious. I'll let Mark and Herman do testing at scale. Reviewers: hermanlee4, MarkCallaghan Reviewed By: MarkCallaghan Differential Revision: https://reviews.facebook.net/D37785 |
4f4a2f6
|
Owner
maykov
commented
Let me know if it still slow.
Owner
Still too slow. See https://gist.github.com/mdcallag/77f33bb27c7e4da8ff2c
…On Tue, Apr 28, 2015 at 2:23 PM, Alexey Maykov <notifications@github.com> wrote:
Let me know if it still slow.
—
Reply to this email directly or view it on GitHub
<https://github.com/MySQLOnRocksDB/mysql-5.6/issues/59#issuecomment-97211721>
.
--
Mark Callaghan
mdcallag@gmail.com
Owner
maykov
commented
It looks like I added disowning for the default column family LRU but not for the rest.
Owner
maykov
commented
Did you try with the per column family setting for the cache size? If so, this results in a creation of a separate lru cache for such families which we do not have an access to. Would you retry without that option? It is fine to provide the rocksdb_block_cache_size global option.
Owner
Yes, shutdown is fast now.
…On Wed, Apr 29, 2015 at 4:28 PM, Alexey Maykov <notifications@github.com> wrote:
Did you try with the per column family setting for the cache size? If so,
this results in a creation of a separate lru cache for such families which
we do not have an access to. Would you retry without that option? It is
fine to provide the rocksdb_block_cache_size global option.
—
Reply to this email directly or view it on GitHub
<https://github.com/MySQLOnRocksDB/mysql-5.6/issues/59#issuecomment-97613585>
.
--
Mark Callaghan
mdcallag@gmail.com
|
|
maykov |
Making shutdowns faster by calling DisownData
…
Summary: As above , see the issue #59 Test Plan: I tested for correcntess by looking at the log file and didn't see anything suspicious. I'll let Mark and Herman do testing at scale. Reviewers: hermanlee4, MarkCallaghan Reviewed By: MarkCallaghan Differential Revision: https://reviews.facebook.net/D37785 |
1bc7073
|
|
|
maykov |
Making shutdowns faster by calling DisownData
…
Summary: As above , see the issue MySQLOnRocksDB#59 Test Plan: I tested for correcntess by looking at the log file and didn't see anything suspicious. I'll let Mark and Herman do testing at scale. Reviewers: hermanlee4, MarkCallaghan Reviewed By: MarkCallaghan Differential Revision: https://reviews.facebook.net/D37785 |
170ad8d
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The problem is the time required for RocksDB to call free on each block in the block cache. Sample stack traces during a slow shutdown are at https://gist.github.com/mdcallag/5ce2d163234ba8cc3bb0
By "too slow" I mean it takes ~30 seconds for RocksDB to shutdown when the block cache has ~15G of data
This might be fixed via "cache_->DisownData();" and see https://reviews.facebook.net/D34977