test: Test health check output for full ratios

Test out of order ratios summary and details
Test various full osd conditions summary and details

Signed-off-by: David Zafman <dzafman@redhat.com>

mon, osd: Add detailed full information for now in the mon

Show ceph health doc output in the correct order

Signed-off-by: David Zafman <dzafman@redhat.com>

mon: Issue warning or error if a full ratio out of order

The full ratios should be in this order: nearfull, backfillfull, full, failsafe full

Signed-off-by: David Zafman <dzafman@redhat.com>

mon: Always fix-up full ratios when specified incorrectly in config

Signed-off-by: David Zafman <dzafman@redhat.com>

mon: Use currently configure full ratio to determine available space

This is a bug that would not adjust available space based on the
currently configured full ratio, but rather the mon_osd_full_ratio
default initial value.

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: check_full_status() remove bogus comment and use equivalent computation

We actually compute kb_used as the kb - kb_avail.  We don't have the
statfs() system call issue of non-privileged f_bavail vs f_bfree.  It
was assumed that used was really like (blocks - f_bfree).  It is not.

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Check whether any OSD is full before starting recovery

Add event RecoveryTooFull to move to NotRecovering state

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Add PG state and flag for too full for recovery

New state machine state NotRecovering
New PG state PG_STATE_RECOVERY_TOOFULL

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Add check_osdmap_full() to check for shard OSD fullness

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Rename backfill_request_* to recovery_request_*

To be used by both recovery and backfill

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Check failsafe full and crash on push/pull

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Revamp injectfull op to support all full states

Use check_* for injectable full checks
Use is_* to just test simple cur_state

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Handle backfillfull_ratio just like nearfull and full

Add BACKFILLFULL as a local OSD cur_state
Notify monitor of this new fullness state

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: For testing full disks add injectfull socket command

Signed-off-by: David Zafman <dzafman@redhat.com>

common: Bump ratio for backfillfull from 85% to 90%

Signed-off-by: David Zafman <dzafman@redhat.com>

common: Remove unused config option osd_recovery_threads

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: too_full_for_backfill() returns ostream for reason

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Fail-safe full is a hard stop even for mds

We can't allow OSD to become non-startable even if mds
could be writing as part of file removals.

Signed-off-by: David Zafman <dzafman@redhat.com>

test: Fix intended test flow and restore nearfull-ratio

This is inconsequential but seems to have always been wrong since original
commit 6cafb0e3e0bc5f992c2483a46cb00e83dca035cc

Signed-off-by: David Zafman <dzafman@redhat.com>

test: Switch from pg to osd for set-*-ratio commands

Testing of 6422e0a220fb3f32ccae50e0c7e52dc9984685c6

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Remove unused argument to clear_queued_recovery

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Increase osd_backfill_retry_interval to 30 seconds

Signed-off-by: David Zafman <dzafman@redhat.com>

ceph-objectstore-tool: cleanup comment

Signed-off-by: David Zafman <dzafman@redhat.com>

osd: Fix log message

Signed-off-by: David Zafman <dzafman@redhat.com>

Merge pull request #12536 from ilc/cls_rgw_minor_refactor

cls/rgw: Clean up the "magic string" usage in the cls layer for RGW.

Reviewed-by: Casey Bodley <cbodley@redhat.com>
Reviewed-by: Adam C. Emerson <aemerson@redhat.com>

Merge pull request #14046 from songbaisen/m7

mgr: optimization some judgment and adjust the debug remove value in register_new_pgs

Reviewed-by: John Spray <john.spray@redhat.com>

mgr: optimization some judgment and adjust the debug remove value in register_new_pgs

Signed-off-by: song baisen <song.baisen@zte.com.cn>

Merge pull request #14267 from yunfeiguan/master

cephfs: fix mount point break off problem after mds switch occured

Reviewed-by: Yan, Zheng <zyan@redhat.com>

Merge pull request #13587 from yanghonggang/master

cephfs: fix write_buf's _len overflow problem

Reviewed-by: John Spray <john.spray@redhat.com>

Merge pull request #13522 from singler/wip-reconnect-client

client/Client.cc: after reset session from MDS - reconnect

Reviewed-by: Yan, Zheng <zyan@redhat.com>

Merge pull request #13466 from jcsp/wip-cephfs-ecoverwrites

Add test for is_hacky_ecoverwrites in cephfs pool checks

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #14207 from jcsp/wip-19401

mds: ignore ENOENT on writing backtrace

Reviewed-by: Yan, Zheng <zyan@redhat.com>

Merge pull request #13579 from ukernel/wip-19022

client: drop cap snaps when auth mds session gets closed

Reviewed-by: John Spray <john.spray@redhat.com>

Merge pull request #14490 from liewegas/wip-no-luminous

mon: add mon_debug_no_require_luminous

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #14553 from guihecheng/rgw_file-revert

rgw_file: fix up potential race condition

verified f23

Merge pull request #14546 from dachary/wip-19628-directory-boot

ceph-disk: enable directory backed OSD at boot time

Reviewed-by: Kefu Chai <kchai@redhat.com>

qa/suites/rados/thrash: do not thrash cluster full without luminous flag

The interface changes with the luminous flag; this will fail.

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

qa/suites/rados: expand other collections with no-require-luminous

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

Merge pull request #14551 from tchaikov/wip-19584

mon/OSDMonitor: transit creating_pgs from pgmap when upgrading

Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #14435 from liewegas/wip-bluestore-deferred

os/bluestore: tune deferred_batch_ops separately for hdd and ssd

Reviewed-by: Igor Fedotov <ifedotov@mirantis.com>

Merge pull request #14489 from liewegas/wip-bluestore-fix-flush

os/bluestore: fix narrow osr->flush() race

mon/OSDMonitor: require_luminous_osds for pg-remap[-items]

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

ceph-disk: enable directory backed OSD at boot time

https://github.com/ceph/ceph/commit/539385b143feee3905dceaf7a8faaced42f2d3c6
introduced a regression preventing directory backed OSD from starting at
boot time.

For device backed OSD the boot sequence starts with ceph-disk@.service
and proceeds to

    systemctl enable --runtime ceph-osd@.service

where the --runtime ensure ceph-osd@12 is removed when the machine
reboots so that it does not compete with the ceph-disk@/dev/sdb1 unit at
boot time.

However directory backed OSD solely rely on the ceph-osd@.service unit
to start at boot time and will therefore fail to boot.

The --runtime flag is selectively set for device backed OSD only.

Fixes: http://tracker.ceph.com/issues/19628
Signed-off-by: Loic Dachary <loic@dachary.org>

rgw_file: add locks to fix potential race

Signed-off-by: Gui Hecheng <guihecheng@cmss.chinamobile.com>

Merge pull request #13859 from jcsp/wip-19204

mds: shut down finisher before objecter

Reviewed-by: Yan, Zheng <zyan@redhat.com>

Merge pull request #14505 from jcsp/wip-19566

mds: fix mgrc shutdown

Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>

Merge pull request #14398 from jcsp/wip-19551

mon: emit cluster log messages on MDS health changes

Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>

qa: add log whitelists for MDS health messages

Now that we send these to the cluster log, we must
whitelist them in the tests that exercise those
unhealthy states.

Fixes: http://tracker.ceph.com/issues/19551
Signed-off-by: John Spray <john.spray@redhat.com>

mon/OSDMonitor: transit creating_pgs from pgmap when upgrading

there could be some pg(s) still being created when we are upgrading to
luminous, and the pools holding them are not changed in the sense of
pg_pool_t::last_change after the upgrade and before we scan for
creating pgs. in that case, the existing update_pending_creatings()
will fail to collect the pgs being created before the upgrade.

with this change, the creating_pgs in pgmap are also used for updating
the OSDMonitor's creating_pgs if it's updated.

but we should stopupdating the pgmap once the upgrade completes. i.e.
stop dispatching MSG_PGSTATS messages to PGMonitor if the quorum and all
osds are luminous.

Fixes: http://tracker.ceph.com/issues/19584
Signed-off-by: Kefu Chai <kchai@redhat.com>

Revert "rgw_file.h: fix potential data race condition"

This reverts commit 3f7b411e1e304e8a6f295dc50b8cf3c6c496f4cc.
Causing hung up due to double lock.

Signed-off-by: Gui Hecheng <guihecheng@cmss.chinamobile.com>

Merge pull request #10851 from atheism/rgw-s3-doc-update

doc/radosgw/s3/cpp.rst: update usage of libs3 APIs to make the examples work

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #14469 from badone/wip-return-error-on-missing-client-id

mds: Return error message instead of asserting

Reviewed-by: John Spray <john.spray@redhat.com>

mon/OSDMonitor: slightly better debug output

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

mon/OSDMonitor: fix creating pg trimming during upgrade

The creating pgs are in the PGMap with the CREATING
state bit set.

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

qa/suites/rados: require_luminous_osds at end of run, then scrub

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

qa/tasks/exec_on_cleanup.py: add

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

Merge pull request #14508 from smithfarm/wip-doc-dev

doc: dev guide: how to run s3-tests locally against vstart

Reviewed-by: Loic Dachary <ldachary@redhat.com>

os/bluestore: tune deferred_batch_ops separately for hdd and ssd

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

Merge pull request #14420 from optimistyzy/329

os/bluestore/NVMEDevice: Add multiple thread support for SPDK I/O thread

Reviewed-by: Haomai Wang <haomai@xsky.com>

doc: dev guide: how to run s3-tests locally against vstart

Add a bunch of verbiage to the Developer Guide

Signed-off-by: Abhishek Lekshmanan <abhishek@suse.com>
Signed-off-by: Nathan Cutler <ncutler@suse.com>

Merge pull request #14458 from linuxbox2/wip-rgw_file-readahead

rgw_file: support readdir cb type hints (plus fixes)

mds: shut down finisher before objecter

Some of the finisher contexts would try to call into Objecter.
We mostly are protected from this by mds_lock+the stopping
flag, but at the Filer level there's no mds_lock, so in the
case of file size probing we have a problem.

Fixes: http://tracker.ceph.com/issues/19204
Signed-off-by: John Spray <john.spray@redhat.com>

Merge pull request #13966 from jcsp/wip-client-logmsg

client: specify inode in get_caps log message

Reviewed-by: Greg Farnum <gfarnum@redhat.com>

Merge pull request #14087 from jcsp/wip-doc-fscid

doc: how to specify filesystem for cephfs clients

Reviewed-by: Yan, Zheng <zyan@redhat.com>

mds: ignore ENOENT on writing backtrace

We get ENOENT when a pool doesn't exist.  This can
happen because we don't prevent people deleting
former cephfs data pools whose files may not have
had their metadata flushed yet.

http://tracker.ceph.com/issues/19401
Signed-off-by: John Spray <john.spray@redhat.com>

rgw_file: don't expire directories being read

If a readdir expire event turns out to be older than last_readdir,
just reschedule it (but actually, we should just discard it, as
another expire event must be in queue.

Fixes: http://tracker.ceph.com/issues/19625
Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

mon: emit cluster log messages on MDS health changes

Previously, when we got a beacon that updated the health
metrics for an MDS, the user would just see mysterious-looking
cluster log messages indicating a rising fsmap epoch number.

It would be good to do this for health messages in general at
some point, but for now just do it for the MDS ones.

Fixes: http://tracker.ceph.com/issues/19551
Signed-off-by: John Spray <john.spray@redhat.com>

mds: fix mgrc shutdown

Were previously only tearing MgrClient down when not
holding a rank, leading to it trying to continue
to run after monclient was shut down.

Fixes: http://tracker.ceph.com/issues/19566
Signed-off-by: John Spray <john.spray@redhat.com>

rgw_file:  chunked readdir

Adjust readdir callback path for new nfs-ganesha chunked readdir,
including changes to respect the result of callback to not
continue.

Pending introduction of offset name hint, our caller will just be
completely enumerating, so it is possible to remove the offset map
and just keep a last offset.

Fixes: http://tracker.ceph.com/issues/19624
Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

rgw_file: introduce rgw_lookup type hints

The new type hints optimize object type deduction, when the
rgw_lookup is called from an rgw_readdir callback.

Fixes: http://tracker.ceph.com/issues/19623
Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

Merge pull request #14468 from tchaikov/wip-show-me-fake-plugin

qa/workunits/ceph-helpers: display rejected string

Reviewed-by: Loic Dachary <ldachary@redhat.com>
Reviewed-by: Brad Hubbard <bhubbard@redhat.com>

mon/OSDMonitor: fix initial map when require_luminous_osds not set on mkfs

If we don't set the luminous flag, we should not set the new luninous
fields or else we'll get a crc mismatch.  (Funnily that happens in the
epoch where the flag is eventually set and the encoded map finally includes
the field we have set in memory.)

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

mon/OSDMonitor: ignore health checks about require_luminous if debug flag is set

This makes it tedious for teuthology health checks to proceed when we
deliberately run luminous osds without this flag.

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

cephfs: fix write_buf's _len overflow problem

After I have set about 400 64KB xattr kv pair to a file,
mds is crashed. Every time I try to start mds, it will crash again.
The root reason is write_buf._len overflowed when doing
Journaler::append_entry().

This patch try to fix this problem through the following changes:

 1. limit file/dir's xattr size
 2. throttle journal entry append operations

Fixes: http://tracker.ceph.com/issues/19033
Signed-off-by: Yang Honggang joseph.yang@xtaotech.com

Merge pull request #14234 from jcsp/wip-19406-dirhash

tools/cephfs: set dir_layout when injecting inodes

Reviewed-by: Yan, Zheng <zyan@redhat.com>

Merge pull request #14472 from joscollin/wip-warning-test

test: add explicit braces to avoid ambiguous ‘else’ and to silence warnings

Reviewed-by: Adam C. Emerson <aemerson@redhat.com>
Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #14495 from tchaikov/wip-cmake-ctags-without-repo

cmake: do not try to add submodule to exclude list if .git is not around

Reviewed-by: Brad Hubbard <bhubbard@redhat.com>

Merge pull request #14404 from tchaikov/wip-another-mgr-command

mon,mgr: move reweight-by-* to mgr

Reviewed-by: Sage Weil <sage@redhat.com>

cmake: do not try to add submodule to exclude list if .git is not around

Signed-off-by: Kefu Chai <kchai@redhat.com>

Merge pull request #14445 from dillaman/wip-rbd-import-diff

rbd: import-diff should discard any zeroed extents

Reviewed-by: Mykola Golub <mgolub@mirantis.com>

Merge pull request #14474 from Wilhelmshaven/fix_non-exist_string

rgw: avoid using null pointer in rgw_file.cc

mds: Return error message instead of asserting

For "session evict" admin socket command return an error message when we
receive an invalid/missing client_id parameter rather than asserting.

Signed-off-by: Brad Hubbard <bhubbard@redhat.com>

mon/OSDMonitor: add mon_debug_no_require_luminous

Add debug option not to set the require_luminous osdmap flag on mkfs, for
debugging purposes.

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

Merge pull request #14482 from trociny/wip-qa-diff

qa/workunits/rbd: diff.sh failed removing nonexistent file

Reviewed-by: Jason Dillaman <dillaman@redhat.com>

tools/cephfs: set dir_layout when injecting inodes

When we left this as zero, the MDS would interpret it was HASH_LINUX
rather than the default HASH_RJENKINS.  Potentially that
could cause problems if there perhaps were already dirfrags in
the metadata pool that were set up using rjenkins.  Mainly
it just seems more appropriate to explicitly set this field
rather than hit the fallback behaviour.

Related: http://tracker.ceph.com/issues/19406
Signed-off-by: John Spray <john.spray@redhat.com>

Merge pull request #14463 from trociny/wip-19451

pybind/rbd: add image metadata methods

Reviewed-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #14471 from trociny/wip-19588

librbd: fix rbd_metadata_list and rbd_metadata_get

Reviewed-by: Jason Dillaman <dillaman@redhat.com>

os/bluestore: fix narrow osr->flush() race

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

Merge pull request #14484 from dillaman/wip-librados-test-stub

test/librados_test_stub: fixed cls_cxx_map_get_keys/vals return value

Reviewed-by: Mykola Golub <mgolub@mirantis.com>

Merge pull request #14375 from cbodley/wip-rgw-fetch-versioned-obj

qa/suites: add mon-reweight-min-pgs-per-osd = 4

so we can run reweight commands in a small cluster with 3 OSD and 27
PGs.

Signed-off-by: Kefu Chai <kchai@redhat.com>

Merge pull request #14476 from wjwithagen/wip-wjw-freebsd-ceph-disk-call-rc.d

ceph-disk/ceph_disk/main.py: fix calling of the bsdrc init scripts

Reviewed-by: Loic Dachary <ldachary@redhat.com>

test/librados_test_stub: fixed cls_cxx_map_get_keys/vals return value

Fixes: http://tracker.ceph.com/issues/19597
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #14399 from ifed01/wip-bluestore-better-better-reuse

os/bluestore: refactor small write handling to reuse blob more effect…

Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #14422 from tchaikov/wip-19427

mon/MonClient: make get_mon_log_message() atomic

Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #14439 from adamemerson/wip-warning-sign

misc: Warning Elimination

Reviewed-by: Kefu Chai <kchai@redhat.com>

qa/workunits/rbd: diff.sh failed removing nonexistent file

Signed-off-by: Mykola Golub <mgolub@mirantis.com>

Merge pull request #14347 from ukernel/wip-19501

mds: make C_MDSInternalNoop::complete() delete 'this'

Reviewed-by: John Spray <john.spray@redhat.com>

bluestore, NVMEDevice: Add multiple thread support for SPDK I/O thread

Previously, we only have one thread to do SPDK I/O, and this patch
adds the multiple thread support.

In this first version, we use the tid of the thread to map the I/Os
of this thread to the corresponding SPDK I/O thread.

Signed-off-by: Ziye Yang <optimistyzy@gmail.com>

rbd: import-diff should discard any zeroed extents

Sparse (zeroed) extents cannot be safely skipped. Instead, the
zeroed extent should be discarded from the image to ensure
the import remains consistent with the export.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>