Adam C. Emerson [Wed, 13 Apr 2022 02:10:31 +0000 (22:10 -0400)]
rgw: Disentangle init_sync_status from RemoteBucketManager
RGWRemoteBucketManager's current design isn't really compatible with
what we need for bucket sync run to work as the number of shards
changes from run to run.
We can make a smaller 'hold information common to all three
operations' class and simplify things a bit.
We also need to fetch `rgw_bucket_index_marker_info` and supply it to
`InitBucketFullSyncStatusCR` to ensure we have the correct generation
and shard count.
Signed-off-by: Adam C. Emerson <aemerson@redhat.com>
Casey Bodley [Fri, 25 Mar 2022 21:14:05 +0000 (17:14 -0400)]
rgw multisite: resharding scales up shard counts 4x faster
in multisite reshard, we need to keep the old index shards around until
other zones finishing syncing from them. we don't want to allow a
bunch of reshards in a row, because we have to duplicate that many
sets of index objects. so we impose a limit of 4 bilog generations (or 3
reshards), and refuse to reshard again until bilog trimming catches up/
trims the oldest generation
under a sustained write workload, a bucket can fill quickly and need
successive reshards. if we have a limit of 3, we should make them count!
so instead of doubling the shard count at each step, multiply by 8
instead when we're in a multisite configuration
Yuval Lifshitz [Wed, 23 Feb 2022 15:21:10 +0000 (17:21 +0200)]
rgw: prevent spurious/lost notifications in the index completion thread
this was happening when asyn completions happened during reshard.
more information about testing:
https://gist.github.com/yuvalif/d526c0a3a4c5b245b9e951a6c5a10517
we also add more logs to the completion manager.
should allow finding unhandled completions due to reshards.
Yuval Lifshitz [Thu, 10 Feb 2022 16:12:55 +0000 (18:12 +0200)]
rgw: refrash the generation of the bucket shard when fetching info
when RGWRados::block_while_resharding() fails because reshard is in
progress, in the next iteration we should fetch the bucket shard
generation. for the case that the generation changed in the middle.
Casey Bodley [Wed, 9 Feb 2022 21:55:38 +0000 (16:55 -0500)]
rgw: prevent 'radosgw-admin bucket reshard' if zonegroup reshard is disabled
dynamic reshard was gated behind the zonegroup resharding flag with
RGWSI_Zone::can_reshard(), but manual reshard was only calling
RGWBucketReshard::can_reshard()
Casey Bodley [Wed, 19 Jan 2022 01:39:37 +0000 (20:39 -0500)]
rgw: RGWBucket::sync() no longer duplicates datalog/bilog entries
RGWSI_BucketIndex_RADOS::handle_overwrite() is already writing the
datalog/bilog entries related to BUCKET_DATASYNC_DISABLED
RGWBucket::sync() calls handle_overwrite() indirectly from
bucket->put_info() when it writes the bucket instance with this new
BUCKET_DATASYNC_DISABLED flag, so RGWBucket::sync() shouldn't
duplicate those writes here
Casey Bodley [Tue, 18 Jan 2022 21:43:42 +0000 (16:43 -0500)]
rgw: use get_current_index() instead of log_to_index_layout()
several places were getting the current index layout indirectly
with layout.logs.back() and rgw::log_to_index_layout(). use
get_current_index() instead so we don't rely on layout.logs, which may
be empty for indexless buckets
rgw: save bucket instance xattrs when resharding cancelled
There appears to be a long-standing bug in RGW such that when
resharding is cancelled and the bucket instance is updated to reflect
the new resharding status, the xattrs were lost. The xattrs are used
to store metadata such as ACLs and LifeCycle policies.
This commit makes sure that all call paths that lead to a cancelled
reshard provide the xattrs, so they can be included when the bucket
instance info is updated.
Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>
rgw: resharding causes bucket attributes to be lost
With the new resharding code, some bucket metadata that is stored as
xattrs (e.g., ACLs, life-cycle policies) were not sent with the
updated bucket instance data when resharding completed. As a result,
resharding has a regression where that metadata is lost after a
successful reshard.
This commit restores the variable in the RGWBucketReshard class that
maintains the bucket attributes, so they can be saved when the bucket
instance object is updated.
Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>
J. Eric Ivancich [Mon, 17 Jan 2022 21:14:16 +0000 (16:14 -0500)]
rgw: add indexless bucket logic to "bucket radoslist"
The "bucket radoslist" sub-command of radosgw-admin is supposed to
list all rados objects tied to one or all directories and thereby
provide a way to determine orphaned rados objects.
But indexless buckets don't provide an index to employ for this
purpose. So warnings or errors should be provided depending on the
circumstances.
Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>
J. Eric Ivancich [Mon, 24 Jan 2022 21:10:57 +0000 (16:10 -0500)]
rgw: update indexless bucket check for bucket stats
The code for bucket stats was recently updated to check for an
indexless bucket before proceeding. The interface on RGWBucketInfo was
recently expanded to support these types of checks, so it is now used.
Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>
J. Eric Ivancich [Mon, 24 Jan 2022 21:08:01 +0000 (16:08 -0500)]
rgw: add streamlined ways to handle indexless buckets correctly
Determining whether a bucket is indexless starting with an
RGWBucketInfo object requires traversing multiple data structures and
"inside knowledge" blurring the line between interface and
implementation. The same applies for retrieving the current index for
non-indexless buckets.
This commit adds to the RGWBucketInfo interface to make this
information readily accessible.
Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>
J. Eric Ivancich [Thu, 23 Dec 2021 21:25:26 +0000 (16:25 -0500)]
rgw: `radosgw-admin bucket stats` on indexless bucket crashes
The new bucket layout code didn't check whether the bucket is
indexless prior to asking for the last entry in the layout log. The
layout log appears to be empty for an indexless bucket, thereby
putting the runtime in an undefined state that later may cause a
failed assertion.
This commit adds two safety checks and returns -EINVAL along with
putting useful information on stderr when either stats are requested
on an indexless bucket or when the layout log is empty.
Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>
Casey Bodley [Mon, 22 Nov 2021 18:05:40 +0000 (13:05 -0500)]
rgw/multisite: RunBucketSourcesSync no longer takes optional target
RGWDataSyncSingleEntryCR is the only caller of RGWRunBucketSourcesSyncCR
it always provides a source_bs, and never provides a target_bs. so remove
all the complexity related to target_bs, and the idea that we'd need to
sync several source bucket shards related to the target bucket
we now just have the single loop over the target buckets that use the
given bucket as a source
Casey Bodley [Mon, 18 Oct 2021 18:15:33 +0000 (14:15 -0400)]
rgw: fix for uninitialized oldest_gen/latest_gen
when data sync queries RGWOp_BILog_Info from an un-upgraded gateway, it
doesn't include the oldest_gen/latest_gen fields. so initialize these
variables to 0 by default
Casey Bodley [Thu, 19 Aug 2021 21:21:37 +0000 (17:21 -0400)]
radosgw-admin: bucket sync status handles missing full status
if the full sync status object is missing, it's possible that we just
haven't started syncing it again after upgrading from just the per-shard
status objects
in this case, as long as we have a log generation 0, assume that we just
haven't initialized the full status object and try to read the gen=0
per-shard incremental status for comparison
Casey Bodley [Thu, 19 Aug 2021 20:12:21 +0000 (16:12 -0400)]
rgw: rgw_read_bucket_inc_sync_status doesn't need bucket info
all we need to construct the per-shard bucket sync status object names
are the bucket names themselves, which we already have from
rgw_sync_bucket_pipe
Casey Bodley [Thu, 19 Aug 2021 20:21:55 +0000 (16:21 -0400)]
rgw: RGWOp_BILog_Status reads full status unconditionally
the calls to rgw_read_bucket_inc_sync_status() depend on
sync_status.incremental_gen, which we need to read via
rgw_read_bucket_full_sync_status() regardless of whether
we're returning it to the client (version > 1)
Casey Bodley [Wed, 9 Jun 2021 16:26:54 +0000 (12:26 -0400)]
rgw: remove destination shard id from rgw_bucket_sync_pair_info
the sync_pair is used as input to RGWBucketPipeSyncStatusManager::status_oid()
to generate the per-shard sync status object names
this sync status tracks incremental bucket sync, which reads changes
from a source bucket's bilog shard, and copies objects from the remote
source bucket to the local destination bucket
this doesn't require sync to know anything about the destination bucket
shards, so rgw_bucket_sync_pair_info and status_oid() now only track the
the destination's rgw_bucket instead of rgw_bucket_shard
Casey Bodley [Tue, 23 Feb 2021 19:55:50 +0000 (14:55 -0500)]
rgw: add custom json encode/decode for the v1 notify API
this adds wrapper structs rgw_data_notify_v1_encoder and
rgw_data_notify_v1_decoder that can encode/decode the v1 json format
directly on the v2 data structure
projects / ceph.git / log