From: Wang Shilong <wshilong@xxxxxxx> Currently it has been popular that single disk could be more than TiB, etc 16Tib with only one single disk, with this trend, one single filesystem could be larger and larger and easily reach PiB with LUN system. The journal filesystem like ext4 need be offline to do regular check and repair from time to time, however the problem is e2fsck still do this using single thread, this could be challenging at scale for two reasons: 1) even with readahead, IO speed still limits several tens MiB per second. 2) could not utilize CPU cores. It could be challenging to try multh-threads for all phase of e2fsck, but as first step, we might try this for most time-consuming pass1, according to our benchmarking it cost of 80% time for whole e2fck phase. Pass1 is trying to scanning all valid inode of filesystem and check it one by one, and the patchset idea is trying to split these to different threads and trying to do this at the same time, we try to merge these inodes and corresponding inode's extent information after threads finish. To simplify complexity and make it less error-prone, the fix is still serialized, since most of time there will be only minor errors for filesystem, what's important for us is parallel reading and checking. Here is a benchmarking on our Lustre filesystem with 1.2 PiB OSD ext4 based filesystem: DDN SFA18KE StorageServer DCR(DeClustering RAID) with 162 x HGST 10TB NL-SAS Tested Server A Virtual Machine running on SFA18KE 8 x CPU cores (Xeon(R) Gold 6140) 150GB memory CentoOS7.7 (Lustre patched kernel) Created 600 Million x 32K byte files. Without Patch With Patch thr=64 pass1: 13079.66 488.57 seconds Total: 15673.33 3188.42 We have 5x total time reduction of total time which is very inspiring. I've tested the whole patch series using 'make test' of e2fsck itself, and i manually set default threads to 4 which still pass almost of test suite, failure cases are below: f_h_badroot f_multithread f_multithread_logfile f_multithread_no h_h_badroot failed because out of order checking output, and others are because of extra multiple threads log output. So the whole series is reasonably stable if you are intrested testing it on different platforms, i've pushed it to github: https://github.com/wangshilong/e2fsprogs pfsck_pass1_v1 It is definitely in early stage, but i'd like to send it for early review for any comments or testing etc. Thanks you very much! Shilong Li Dongyang (1): libext2fs: optimize ext2fs_convert_subcluster_bitmap() Li Xi (25): e2fsck: add -m option for multithread e2fsck: copy context when using multi-thread fsck e2fsck: copy fs when using multi-thread fsck e2fsck: copy dblist when using multi-thread fsck e2fsck: clear icache when using multi-thread fsck e2fsck: add assert when copying context e2fsck: copy bitmaps when copying context e2fsck: copy badblocks when copying fs e2fsck: open io-channel when copying fs e2fsck: create logs for mult-threads e2fsck: create one thread to fsck e2fsck: add start/end group for thread e2fsck: split groups to different threads e2fsck: print thread log properly e2fsck: merge bitmaps after thread completes e2fsck: do not change global variables e2fsck: optimize the inserting of dir_info_db e2fsck: merge dir_info after thread finishes e2fsck: rbtree bitmap for dir e2fsck: merge badblocks after thread finishes e2fsck: merge icounts after thread finishes e2fsck: merge dblist after thread finishes e2fsck: add debug codes for multiple threds e2fsck: merge counts when threads finish LU-8465 e2fsck: merge fs flags when threads finish Wang Shilong (20): e2fsck: cleanup struct e2fsck_struct e2fsck: merge dx_dir_info e2fsck: make threads splitting aware of flex_bg e2fsck: merge dirs_to_hash when threads finish e2fsck: merge context flags properly e2fsck: split and merge quota context e2fsck: serialize fix operations e2fsck: move some fixes out of parallel pthreads e2fsck: split and merge invalid bitmaps e2fsck: fix to protect EA checking e2fsck: allow admin specify number of threads e2fsck: kickoff mutex lock for block found map e2fsck: fix readahead for pfsck of pass1 e2fsck: kick off ea mutex lock from pfsck e2fsck: merge encrypted_files after threads finish e2fsck: merge inode_bad_map after threads finish e2fsck: simplify e2fsck context merging codes e2fsck: merge options after threads finish e2fsck: reset lost_and_found after threads finish LU-8465 e2fsck: merge extent depth count after threads finish configure.ac | 6 + e2fsck/dirinfo.c | 220 ++- e2fsck/dx_dirinfo.c | 67 + e2fsck/e2fsck.h | 336 +++-- e2fsck/encrypted_files.c | 175 ++- e2fsck/logfile.c | 12 +- e2fsck/pass1.c | 1692 ++++++++++++++++++++--- e2fsck/problem.c | 9 + e2fsck/problem.h | 3 + e2fsck/unix.c | 33 +- e2fsck/util.c | 56 +- lib/ext2fs/badblocks.c | 75 +- lib/ext2fs/bitmaps.c | 8 + lib/ext2fs/bitops.h | 2 + lib/ext2fs/blkmap64_rb.c | 51 + lib/ext2fs/bmap64.h | 3 + lib/ext2fs/dblist.c | 36 + lib/ext2fs/ext2_err.et.in | 3 + lib/ext2fs/ext2_io.h | 2 + lib/ext2fs/ext2fs.h | 11 + lib/ext2fs/ext2fsP.h | 1 - lib/ext2fs/gen_bitmap64.c | 89 +- lib/ext2fs/icount.c | 102 ++ lib/ext2fs/undo_io.c | 19 + lib/ext2fs/unix_io.c | 24 +- lib/support/mkquota.c | 19 + lib/support/quotaio.h | 2 + tests/f_itable_collision/expect.1 | 3 - tests/f_multithread/expect.1 | 25 + tests/f_multithread/expect.2 | 7 + tests/f_multithread/image.gz | 1 + tests/f_multithread/name | 1 + tests/f_multithread/script | 4 + tests/f_multithread_completion/expect.1 | 2 + tests/f_multithread_completion/expect.2 | 23 + tests/f_multithread_completion/image.gz | 1 + tests/f_multithread_completion/name | 1 + tests/f_multithread_completion/script | 4 + tests/f_multithread_logfile/expect.1 | 25 + tests/f_multithread_logfile/image.gz | 1 + tests/f_multithread_logfile/name | 1 + tests/f_multithread_logfile/script | 32 + tests/f_multithread_no/expect.1 | 26 + tests/f_multithread_no/expect.2 | 23 + tests/f_multithread_no/image.gz | 1 + tests/f_multithread_no/name | 1 + tests/f_multithread_no/script | 4 + tests/f_multithread_preen/expect.1 | 11 + tests/f_multithread_preen/expect.2 | 23 + tests/f_multithread_preen/image.gz | 1 + tests/f_multithread_preen/name | 1 + tests/f_multithread_preen/script | 4 + tests/f_multithread_yes/expect.1 | 2 + tests/f_multithread_yes/expect.2 | 23 + tests/f_multithread_yes/image.gz | 1 + tests/f_multithread_yes/name | 1 + tests/f_multithread_yes/script | 4 + 57 files changed, 2858 insertions(+), 455 deletions(-) create mode 100644 tests/f_multithread/expect.1 create mode 100644 tests/f_multithread/expect.2 create mode 120000 tests/f_multithread/image.gz create mode 100644 tests/f_multithread/name create mode 100644 tests/f_multithread/script create mode 100644 tests/f_multithread_completion/expect.1 create mode 100644 tests/f_multithread_completion/expect.2 create mode 120000 tests/f_multithread_completion/image.gz create mode 100644 tests/f_multithread_completion/name create mode 100644 tests/f_multithread_completion/script create mode 100644 tests/f_multithread_logfile/expect.1 create mode 120000 tests/f_multithread_logfile/image.gz create mode 100644 tests/f_multithread_logfile/name create mode 100644 tests/f_multithread_logfile/script create mode 100644 tests/f_multithread_no/expect.1 create mode 100644 tests/f_multithread_no/expect.2 create mode 120000 tests/f_multithread_no/image.gz create mode 100644 tests/f_multithread_no/name create mode 100644 tests/f_multithread_no/script create mode 100644 tests/f_multithread_preen/expect.1 create mode 100644 tests/f_multithread_preen/expect.2 create mode 120000 tests/f_multithread_preen/image.gz create mode 100644 tests/f_multithread_preen/name create mode 100644 tests/f_multithread_preen/script create mode 100644 tests/f_multithread_yes/expect.1 create mode 100644 tests/f_multithread_yes/expect.2 create mode 120000 tests/f_multithread_yes/image.gz create mode 100644 tests/f_multithread_yes/name create mode 100644 tests/f_multithread_yes/script -- 2.25.2
