bluestore read regression testing

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



This is mostly for Igor though I thought I'd post to ceph-devel if anyone else trying to track down the bluestore read regression.

Here's what I'm seeing so far after doing some bisecting tonight. Sadly it's slow going since we've got a lot of skip steps due to broken builds/etc. I'm testing my own wip branch with master from last week plus Igor's write path segfault fix here:

https://github.com/markhpc/ceph/tree/wip-bluestore-blobmap-lock

With the latest commit in that branch I'm seeing about 10MB/s for 4K random reads vs 141MB/s for the same test on commit 55cb608. This test is setup with lots of concurrency: 8 fio processes each targeting a 64GB RBD image with an iodepth of 32. The test cluster has 4 nodes with 4 P3700 cards split up into 16 OSDs.

The bisect hasn't made much progress.  This is all I've gotten so far:

[nhm@espresso ceph]$ git bisect log
git bisect start
# bad: [c7da2f7c869694246650a9276a2b67aed9bf818f] [DNM] Igor's blob_map lock test.
git bisect bad c7da2f7c869694246650a9276a2b67aed9bf818f
# good: [55cb608f63787f7969514ad0d7222da68ab84d88] Merge pull request #9563 from markhpc/wip-rocksdb-jemalloc
git bisect good 55cb608f63787f7969514ad0d7222da68ab84d88
# good: [55cb608f63787f7969514ad0d7222da68ab84d88] Merge pull request #9563 from markhpc/wip-rocksdb-jemalloc
git bisect good 55cb608f63787f7969514ad0d7222da68ab84d88
# skip: [4e911c8c8b3791582c16b4fdfeda7269f751428a] Merge pull request #9821 from dx9/wip-ac-alpine
git bisect skip 4e911c8c8b3791582c16b4fdfeda7269f751428a
# skip: [ef272731215ccf0a4178ae01f2903077488ef895] mds: improve comments for ENOTEMPTY checks
git bisect skip ef272731215ccf0a4178ae01f2903077488ef895
# good: [abd84f4f1b804aa6a90ed5f8b94c1cfddc4e8956] os/filestore: fix statfs
git bisect good abd84f4f1b804aa6a90ed5f8b94c1cfddc4e8956
# skip: [c83374ff717a106cd5638b3df38536b8cd579d87] Merge pull request #9626 from liewegas/wip-bluestore-2q
git bisect skip c83374ff717a106cd5638b3df38536b8cd579d87
# skip: [9dc3eb61175e07f2aca858b431e60bb38719104d] Merge pull request #9890 from xiexingguo/xxg-wip-fix-bfs-alloc
git bisect skip 9dc3eb61175e07f2aca858b431e60bb38719104d
# skip: [de6b2584b0478ec7b11f0e9bb31d29a0a6a71cad] os/bluestore: switch to intrusive_set for blobs
git bisect skip de6b2584b0478ec7b11f0e9bb31d29a0a6a71cad

that yields a whole lot of commits that still need to be looked through given the number of skips I'm hitting.

[nhm@espresso ceph]$ git log --pretty=oneline abd84f4f1..c7da2f7c8 | wc
   1165    8728  110448

Anyway, I'll have more tomorrow, but that's all I've got tonight. Happy hunting!

Mark


--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html



[Index of Archives]     [CEPH Users]     [Ceph Large]     [Information on CEPH]     [Linux BTRFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]
  Powered by Linux