Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

rgw: split logs for metadata sync by period #7699

Closed
wants to merge 693 commits into from
Closed

rgw: split logs for metadata sync by period #7699

wants to merge 693 commits into from

Conversation

cbodley
Copy link
Contributor

@cbodley cbodley commented Feb 18, 2016

This is a first draft of the metadata sync refactor that separates the logs for each period. It passes the existing multisite tests, but those only create a single period so don't exercise most of the new logic.

TODO: implement find_oldest_log_period() for RGWMetadataManager, which is currently a hack that returns the current period
TODO: update multisite tests to operate over multiple periods

yehudasa and others added 30 commits October 30, 2015 17:03
@yehudasa
rgw: move children collection to other method
65ba101 
and call it again instead of just draining at the end

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
@yehudasa
rgw: rework incremental md sync error handling
9ebc90a 
similar to what we do with the full md sync. Identify transient
errors, and if so return -EAGAIN so that caller would retry.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
@yehudasa
rgw: move data sync init into the processing thread
f15b326 
that way we don't depend on remote peer to be available

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
@yehudasa
rgw: fix data sync initialization
67e11cc 
Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
@oritwas
rgw: allow creating a bucket without current period
1e91c95 
Signed-off-by: Orit Wasserman <owasserm@redhat.com>
@oritwas
rgw: use default zonegroup and zone in init_complete
b9ca86a 
Signed-off-by: Orit Wasserman <owasserm@redhat.com>
@oritwas
rgw: fix bucket creation and deletion when there is no realm or perio…
8836a77 
…d configured

Signed-off-by: Orit Wasserman <owasserm@redhat.com>
@cbodley
rgw: realm watcher locks mutex for SafeTimer::shutdown()
e196cb6 
this fixes an assert() on shutdown

Signed-off-by: Casey Bodley <cbodley@redhat.com>
@oritwas
rgw: bucket ops, obj create and obj delete should check to see if buc…
62f75c7 
…ket exists

rgw_build_policies set the bucket_exists flag

Signed-off-by: Orit Wasserman <owasserm@redhat.com>
@oritwas
rgw: workaround to clear location constrain when using default zonegroup
ebaef77 
Signed-off-by: Orit Wasserman <owasserm@redhat.com>
@cbodley
rgw: RGWRestConn uses vector instead of map
2df1d39 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: move RGWPostRESTResourceCR to rgw_cr_rest.h
5f02d46 
for use by RGWRealmPusher

Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: use insert to copy containers in rgw_rest_conn
d7f7b49 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: add typedef for rest parameter lists
c81cc28 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: add make_param_list() helper
651f5cb 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: RGWRESTPostResource takes param_list_t for RGWPostRESTResourceCR
565f64c 
RGWPostRESTResourceCR needs to make a copy of the parameters given to
its constructor, because it doesn't use them until send_request(). this
results in a segfault when the given parameter list is on the stack

Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: apply RGWRESTReadResource::wait_bl() fix to PostResource as well
4d3e73b 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: use boost::intrusive_ptr for http_op ref counting
d1b1f9f 
this fixes a segfault in the RGWPostRESTResourceCR::request_complete()
error path, which accessed http_op after calling put()

Signed-off-by: Casey Bodley <cbodley@redhat.com>
@yehudasa
Merge pull request ceph#6480 from cbodley/wip-rgw-rest-conn
897e056 
[rgw multisite] refactoring and fixes for RGWREST

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>
@yehudasa
rgw: shutdown lease CR on mdlog inc sync failure
5cfaa2b 
Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
@yehudasa
rgw: update the correct shards when doing full sync
7f8461b 
The shard we use for each entry (either meta, or data) needs to be selected
correctly to match the different logs we keep. Otherwise async notifications
don't work correctly

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
@yehudasa
rgw: keep sync progress info for full sync
e0e185b 
Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
@yehudasa
rgw: show data full sync stats
1414313 
Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
@yehudasa
Merge pull request ceph#6433 from cbodley/wip-rgw-period-sync-status
e1e8d99 
rgw: store metadata sync status on period commit

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>
@oritwas
rgw: remove include to <boost/asio/coroutine.h>
e1886c7 
Signed-off-by: Orit Wasserman <owasserm@redhat.com>
@oritwas
rgw: remove include to <boost/asio/coroutine.h>
e21cf98 
Signed-off-by: Orit Wasserman <owasserm@redhat.com>
@oritwas
rgw: fix compilation with older boost versions that don't have asio c…
9ce4d81 
…oroutines

Signed-off-by: Orit Wasserman <owasserm@redhat.com>
@cbodley
rgw: coroutine manager returns ECANCELED when going down
6372451 
RGWRemoteMetaLog::run_sync() was getting stuck in an infinite loop after
a call to RGWCoroutinesManager::stop().  instead of returning success
when we see 'going_down', fail the coroutine with ECANCELED

the other RGWCoroutinesManager::run() overload also needed modification
to prevent us from overwriting this error code

Signed-off-by: Casey Bodley <cbodley@redhat.com>
@yehudasa
Merge pull request ceph#6529 from cbodley/wip-rgw-coroutine-going-down
9f0a266 
rgw: coroutine manager returns ECANCELED when going down

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>
@yehudasa
rgw: init REST connection by zone name
54ade1a 
Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
@cbodley
rgw: remove unused clone_shards() in metadata sync
232f258 
clone_shards() was an exact duplicate for fetch()

Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: remove unused class RGWReadMDLogShardInfo
4e029d8 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: remove unused RGWRemoteMetaLog::list_shards
d3ff605 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: remove RemoteMetaLog::fetch() and admin command
a2ebade 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
cmake: remove duplicate rgw source files
6c7b512 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: use std::move for strings in RGWFetchAllMetaCR
56b37e7 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: include period id in rgw_meta_sync_info
4c3fea6 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: move sync_status into RGWRemoteMetaLog
cd41441 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: RGWRemoteMetaLog passes period id with requests to master
c53a6be 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: add period id to metadata log oids
8446fec 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: add map of period_id -> RGWMetadataLog
6b0fef5 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw-admin: update callers of get_log
e6d92e6 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: add period id to rgw_rest_log operations
f4368b7 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: factor get_log_shard_id() out of RGWMetadataLog
1f4ef0b 
by using RGWRados::key_to_shard_id(), we can avoid the dependency on
RGWMetadataLog's prefix (and period)

Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: add first log period to mdlog info response
2b7c32c 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: rados coroutines take bucket by const ref
6c7df6e 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: remove RGWMetadataManager::store_md_log_entries
6f660ac 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: get period history cursor when starting sync
b12e457 
Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley
rgw: RGWMetaSyncCR loops through period history
c200efc 
RGWMetaSyncCR uses a period history Cursor to track its position. it
uses this to get the max sync markers for each shard from the following
period, so that RGWMetaSyncShardCR knows when to stop syncing and return
control to RGWMetaSyncCR

Signed-off-by: Casey Bodley <cbodley@redhat.com>
@cbodley cbodley added the rgw label Feb 18, 2016
@cbodley
Copy link
Contributor Author

cbodley commented Feb 18, 2016

@yehudasa there are two XXX: comments that I'd appreciate your feedback on, in addition to the design in general

@yehudasa
Copy link
Member

@cbodley ok, thanks. I'll take a look at it.

yehudasa
yehudasa reviewed Feb 18, 2016
View reviewed changes
src/rgw/rgw_sync.cc
@@ -1344,9 +1258,10 @@ class RGWMetaSyncShardCR : public RGWCoroutine {
sync_marker.marker = sync_marker.next_step_marker;
sync_marker.next_step_marker.clear();
}
// XXX: why write the marker if !can_adjust_marker?
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@cbodley I think you have a point here, probably want to skip writing the marker in that case

@yehudasa yehudasa force-pushed the wip-rgw-new-multisite branch 2 times, most recently from 4c4e501 to ac168e3 Compare February 18, 2016 23:24
@yehudasa
Copy link
Member

rebased, merged in a different PR

@yehudasa yehudasa closed this Feb 19, 2016
@cbodley cbodley mentioned this pull request Feb 26, 2016
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
rgw
Projects
None yet
Milestone
No milestone
3 participants
@cbodley @yehudasa @oritwas