rgw multisite: fix the increamtal bucket sync init

[Aran85]

in the RGWBucketShardFullSyncCR::operate, inc_marker will assigned with remote bilog's max_marker.
but the sync_status's inc_marker cant be assigned.so the next step inc sync will always sync
from null log,which means at beginning log.

Signed-off-by: Zengran Zhang zhangzengran@h3c.com

[cbodley]

thanks, this looks correct. i'll run some tests to validate

[cbodley]

@Aran85 I created a ticket for this bug at http://tracker.ceph.com/issues/17624 so we can make sure this gets backported to jewel. could you please add Fixes: http://tracker.ceph.com/issues/17624 to your commit message?

[Aran85]

@cbodley done

[cbodley]

tested successfully using two vstart clusters with one zone each. i started a gateway on the first zone and created a bucket with ~20 objects. i then started the second zone's gateway, and watched it go through full sync. without this fix, the first request for incremental sync uses an empty marker. with the fix, it uses a valid marker 👍

that said, this points out another issue with RGWRunBucketSyncCoroutine - it shouldn't be keeping a cached copy of the sync status outside of the lock. rather than each of the InitBucketShardSyncStatus, BucketShardFullSync, and BucketShardIncrementalSync coroutines taking and releasing that lock, RunBucketSync should acquire it at the beginning and hold it over all of the other calls

RunBucketSync also doesn't appear to be handling the parallel full and incremental sync correctly, so we'll want to get that working for #10995. i'll work on a design that covers both of these issues, and share with the ceph-devel list

[cbodley]

@Aran85 this was causing a segfault in the radosgw-admin bucket sync init command because RGWBucketSyncStatusManager::init_sync_status() was returning a coroutine that had a pointer to a temporary variable. i opened #11594 with the fix

i also opened #11598 that does the initial lease refactoring for RGWRunBucketSyncCoroutine