Sage Weil
0db8c4c773
osd: properly encode ondisklog attr
...
Also combine into a single attr.
2008-12-18 16:01:41 -08:00
Sage Weil
ee4b96187b
buffer: optionally preallocate some append_buffer
2008-12-18 15:24:00 -08:00
Sage Weil
9f1c7b16c1
make time format easier to read
2008-12-18 11:34:11 -08:00
Sage Weil
41dfd4089d
ceph: terminate dumped output
2008-12-17 16:48:38 -08:00
Sage Weil
1a94147b2c
mon: tell stdout where the monmap came from
2008-12-17 15:42:23 -08:00
Sage Weil
705496f660
dstart: cm.txt update (for 3x)
2008-12-17 15:42:22 -08:00
Sage Weil
705918340f
ceph: fix observe renewal; ack by paxos
2008-12-17 15:43:44 -08:00
Sage Weil
901e58266f
todo
2008-12-17 14:06:15 -08:00
Sage Weil
022d80126f
verify-mds-journal: periodically verify mds replay; stop if error
2008-12-17 13:37:34 -08:00
Sage Weil
75421fcdb4
mds: only standby-replay when explicitly requested on startup.
...
Until it does something more than just replay the log and quit.
2008-12-17 13:32:11 -08:00
Sage Weil
6bf3f40190
mds: basic standby-replay hack
...
Replay the log, and exit cleanly if everything looks ok.
2008-12-17 13:12:27 -08:00
Sage Weil
81be48b5b4
mon: replace laggy standby-replay nodes too
...
If there is someone to take over...
2008-12-17 13:02:44 -08:00
Sage Weil
7b695ec689
mon: mark extra standby mds's as standby-replay
...
Not that the mds does the right thing, yet...
2008-12-17 12:34:28 -08:00
Sage Weil
c941da1303
osd: enable queue_generate_backlog on pg query
...
Code was disabled, oops.
2008-12-17 11:21:09 -08:00
Sage Weil
450cfeeec4
mds: fix mdsmap dump
...
Don't fall over duplicate mds (id,inc) pairs.
2008-12-17 11:20:23 -08:00
Sage Weil
ba140b5a6d
msgr: stop accepter on socket error
...
Only if we get a few in a row.
2008-12-17 10:32:09 -08:00
Sage Weil
82b0bc39e5
mds: restructure mds map, clean up mdsmon
...
More flexible mdsmap structure. Cleaner interface.
2008-12-17 10:32:09 -08:00
Sage Weil
6fb760b572
mon: clean up mdsmon a bit; do takeovers from tick(), not _committed()
...
This puts all the takeover logic in one place. And avoids any
funny issues where the monitor fails before it can do a takeover.
More robust.
2008-12-17 10:32:09 -08:00
Sage Weil
80302174fe
mon: blacklist failed mds
2008-12-17 10:32:09 -08:00
Sage Weil
c04ef1a280
mon: expire old blacklist items (osdmap change)
2008-12-17 10:32:09 -08:00
Sage Weil
117fc9d82d
osd: clean out acker cruft
2008-12-17 10:32:09 -08:00
Sage Weil
d2d30489e6
rev protocol, mon disk formats due to mdsmap change
2008-12-17 10:32:09 -08:00
Sage Weil
b3aa3ed7c3
mon: mdsmap infrastructure for standby states, up:standby-replay
...
We need a state value for standby mds's.
2008-12-17 10:32:09 -08:00
Sage Weil
17d5ff6bba
osd: fix pg stat reporting
...
We were skipping pgs that weren't active instead of not primary.
2008-12-16 13:37:35 -08:00
Sage Weil
dd091b38d5
osd: do delayed activation after replay via a queue, not timer event
...
This avoids osd_lock dependency by using osd->timer.
2008-12-16 13:37:32 -08:00
Sage Weil
6d28c8623f
osd: take osd_lock in generate_backlog before peer()
...
Peer() uses the osd.timer to schedule the replay interval, which
needs osd_lock.
2008-12-16 13:18:23 -08:00
Sage Weil
e48fc3c586
mds cleanup
2008-12-16 11:32:44 -08:00
Sage Weil
e1abc7fad3
mds: add logclient
2008-12-16 11:25:21 -08:00
Sage Weil
cff97c1247
logclient: adjust link_dispatcher; add unlink_dispatcher
2008-12-16 11:25:14 -08:00
Sage Weil
b917de3b80
todos
2008-12-16 10:56:53 -08:00
Sage Weil
d0041ee81c
mds: remove follows==0 special cases
...
A follows==0 shouldn't have any special meaning anymore. See also
1876ca5ad4
2008-12-15 16:40:08 -08:00
Sage Weil
dc0cc83a44
mds: when recovering size, don't munge up projected->size; use new_size
2008-12-15 15:57:36 -08:00
Sage Weil
415b4bbca8
mds: update rbytes with size on truncate, etc.
2008-12-15 15:56:51 -08:00
Sage Weil
48c3da005f
mon: send mdsmap on beacon from mds not in the map
...
This happens when an mds goes laggy and is marked down by the
monitor.
2008-12-15 14:33:00 -08:00
Sage Weil
016cf1e477
mon: clean up mds failure output
...
iterator p isn't valid; use temp values.
2008-12-15 14:17:26 -08:00
Sage Weil
2b0337fddf
mon: immediately propose after 'osd setmap'
...
Any subsequent osdmap changes will be ignored anyway.
Note that this still throws out changes _prior_ to the setmap. In
theory, that shouldn't matter, since we're replacing the map
anyway.
2008-12-15 13:44:52 -08:00
Sage Weil
1a5c8ae2b1
update debian, spec files to reflect cmonctl->ceph rename
2008-12-15 12:33:10 -08:00
Sage Weil
6cd31dae26
ceph: allow > and < to redirect command input/output
2008-12-15 12:00:17 -08:00
Sage Weil
483dba8d68
ceph: fold cobserver into ceph
2008-12-15 11:45:10 -08:00
Sage Weil
1aa9446036
todo
2008-12-15 11:34:13 -08:00
Sage Weil
213eae1bc2
mds: mark new directories new in journal; add to new list on replay
...
This ensures the dir is written when the logseg is eventually
expired.
2008-12-15 11:34:13 -08:00
Sage Weil
2cd256fe06
cobserver: usage
2008-12-15 11:34:13 -08:00
Sage Weil
8b0bd3ea17
rename cmonctl -> ceph
2008-12-15 11:34:13 -08:00
Yehuda Sadeh
c75a09989b
cobserver: retry if when no response on startup
2008-12-15 11:21:33 -08:00
Yehuda Sadeh
3e37543c3f
vstart.sh: can specify mon address
2008-12-15 10:45:08 -08:00
Yehuda Sadeh
e1f536d1ae
osd: add missing declaration
2008-12-15 09:54:50 -08:00
Sage Weil
5026d0ee4c
vstart: debug pg reporting
2008-12-14 21:40:21 -08:00
Sage Weil
b3f831f730
osd: generate_backlog asynchronously in a work queue; simplify peering a bit
...
We do all backlog creation in a thread pool. Break it down into the
disk scan and log integration steps, and drop PG lock as much as possible.
We only worry about pg acting changes; backlogs are only generated when the
pg is inactive.
We also simplify the activation code a bit by observing that replicas only
generate backlogs when their logs are discontiguous with the primary; in
such cases, we pull the backlog during peering and no generate_backlog
(equivalent) is needed for activation.
2008-12-14 21:39:54 -08:00
Sage Weil
aab8882613
osd: half-finished backlog_wq
2008-12-12 20:25:03 -08:00
Sage Weil
0aecb46816
crush: don't recurse to leaf unless item is a bucket
...
This avoids choking on 'chooseleaf indep 0 item device' (it's
equivalent to 'choose indep 0 item device').
2008-12-12 20:22:25 -08:00
