Since #34288, we might hit deadlock if the FollowTask has more fetchers
than writers. This can happen in the following scenario:
Suppose the leader has two operations [seq#0, seq#1]; the FollowTask has
two fetchers and one writer.
1. The FollowTask issues two concurrent fetch requests: {from_seq_no: 0,
num_ops:1} and {from_seq_no: 1, num_ops:1} to read seq#0 and seq#1
respectively.
2. The second request which fetches seq#1 completes before, and then it
triggers a write request containing only seq#1.
3. The primary of a follower fails after it has replicated seq#1 to
replicas.
4. Since the old primary did not respond, the FollowTask issues another
write request containing seq#1 (resend the previous write request).
5. The new primary has seq#1 already; thus it won't replicate seq#1 to
replicas but will wait for the global checkpoint to advance at least
seq#1.
The problem is that the FollowTask has only one writer and that writer
is waiting for seq#0 which won't be delivered until the writer completed.
This PR proposes to replicate existing operations with the old primary
term (instead of the current term) on the follower. In particular, when
the following primary detects that it has processed an process already,
it will look up the term of an existing operation with the same seq_no
in the Lucene index, then rewrite that operation with the old term
before replicating it to the following replicas. This approach is
wait-free but requires soft-deletes on the follower.
Relates #34288
Today we rely on the LocalCheckpointTracker to ensure no duplicate when
enabling optimization using max_seq_no_of_updates. The problem is that
the LocalCheckpointTracker is not fully reloaded when opening an engine
with an out-of-order index commit. Suppose the starting commit has seq#0
and seq#2, then the current LocalCheckpointTracker would return "false"
when asking if seq#2 was processed before although seq#2 in the commit.
This change scans the existing sequence numbers in the starting commit,
then marks these as completed in the LocalCheckpointTracker to ensure
the consistent state between LocalCheckpointTracker and Lucene commit.
This change makes it no longer possible to follow / auto follow without
specifying a leader cluster. If a local index needs to be followed
then `cluster.remote.*.seeds` should point to nodes in the local cluster.
Closes#34258
The `AutoFollowTests` needs to restart the clusters between each tests, because
it is using auto follow stats in assertions. Auto follow stats are only reset
by stopping the elected master node.
Extracted the `testGetOperationsBasedOnGlobalSequenceId()` test to its own test, because it just tests the shard changes api.
* Renamed AutoFollowTests to AutoFollowIT, because it is an integration test.
Renamed ShardChangesIT to IndexFollowingIT, because shard changes it the name
of an internal api and isn't a good name for an integration test.
* move creation of NodeConfigurationSource to a seperate method
* Fixes issues after merge, moved assertSeqNos() and assertSameDocIdsOnShards() methods from ESIntegTestCase to InternalTestCluster, so that ccr tests can use these methods too.
Today we rewrite the operations from the leader with the term of the
following primary because the follower should own its history. The
problem is that a newly promoted primary may re-assign its term to
operations which were replicated to replicas before by the previous
primary. If this happens, some operations with the same seq_no may be
assigned different terms. This is not good for the future optimistic
locking using a combination of seqno and term.
This change ensures that the primary of a follower only processes an
operation if that operation was not processed before. The skipped
operations are guaranteed to be delivered to replicas via either
primary-replica resync or peer-recovery. However, the primary must not
acknowledge until the global checkpoint is at least the highest seqno of
all skipped ops (i.e., they all have been processed on every replica).
Relates #31751
Relates #31113
Also fixed ShardFollowNodeTaskTests to not return ops when responseSize
is empty. Otherwise ops are returned when no ops are expected to be returned.
Co-authored-by: Jason Tedor <jason@tedor.me>
Unfollow should be allowed / disallowed on a per index level instead of
cluster level.
Also renamed `create_follow_index` index privilege to
`manage_follow_index` privilege and include unfollow and close APIs.
This commit modifies the follow stats API response structure to more
clearly highlight meaning of the higher level fields. In particular,
previously the response had a top-level key for each index. Instead, we
nest the indices under an "indices" field which is now an array. The
values in this array are objects containing two fields: "index" which is
the name of the follower index, and "shards" which is an array where
each value in the array is the follower stats for that shard. That is,
we have gone from:
{
"bar": [
{
"shard_id": 0...
}...
]...
}
to
{
"indices": [
{
"index": "bar",
"shards": [
{
"shard_id": 0...
}...
]
}...
}
In the CCR docs we want to refer to the endpoint that returns following
stats as the follow stats API. This commit renames the internal
implementation of this endpoint to reflect this usage.
The follower index shard history UUID will be fetched from the indices stats api when the shard follow task starts and will be provided with the bulk shard operation requests. The bulk shard operations api will fail if the provided history uuid is unequal to the actual history uuid.
No longer record the leader history uuid in shard follow task params, but rather use the leader history UUIDs directly from follower index's custom metadata. The resume follow api will remain to fail if leader index shard history UUIDs are missing.
Closes#33956
Since #34099, the FollowingEngine will skip an operation which was
already processed before. With that change, it should be okay to unmute
testFollowIndexAndCloseNode.
This arose when two commits were pushed at roughly the same time, both
of which compiled successfully against master, but not when taken
together. This commit fixes a reference in one of the commits that was
changed in the other commit.
This commit modifies the CCR stats endpoint for indices to be
/{index}/_ccr/stats. This makes this endpoint consistent with other
index-centric endpoints like indices stats.
The unfollow API changes a follower index into a regular index, so that it will accept write requests from clients.
For the unfollow api to work the index follow needs to be stopped and the index needs to be closed.
Closes#33931
This change introduces the indexing optimization using sequence numbers
in the FollowingEngine. This optimization uses the max_seq_no_updates
which is tracked on the primary of the leader and replicated to replicas
and followers.
Relates #33656
Prior to following an index in the follow API, check whether current
user has sufficient privileges in the leader cluster to read and
monitor the leader index.
Also check this in the create and follow API prior to creating the
follow index.
Also introduced READ_CCR cluster privilege that include the minimal
cluster level actions that are required for ccr in the leader cluster.
So a user can follow indices in a cluster, but not use the ccr admin APIs.
Closes#33553
Co-authored-by: Jason Tedor <jason@tedor.me>
The following changes were made:
* Added ElasticsearchSecurityException. For in the case the current user has insufficient privileges while an index is being followed. Prior to following ccr checks whether the current user has sufficient privileges and if not the follow api fails with an error.
* Added Index block exception. If the leader index gets closed, this exception is returned.
* Added ClusterBlockException service unavailable. In case for example the leader cluster is without elected master.
* Removed IndexNotFoundException. If the leader / follower index has been deleted, ccr will need to stop the shard follow tasks with an error.
Closes#33954
* Renamed CCR APIs
Renamed:
* `/{index}/_ccr/create_and_follow` to `/{index}/_ccr/follow`
* `/{index}/_ccr/unfollow` to `/{index}/_ccr/pause_follow`
* `/{index}/_ccr/follow` to `/{index}/_ccr/resume_follow`
Relates to #33931
always use `IndicesOptions.strictExpand()` for indices options.
The follow index may be closed and we still want to get stats from
shard follow task and the whether the provided index name matches with
follow index name is checked when locating the task itself in the ccr
stats transport action.
This commit replicates the max_seq_no_of_updates on the leading index
to the primaries of the following index via ShardFollowNodeTask. The
max_seq_of_updates is then transmitted to the replicas of the follower
via replication requests (that's BulkShardOperationsRequest).
Relates #33656
We start tracking max seq_no_of_updates on the primary in #33842. This
commit replicates that value from a primary to its replicas in replication
requests or the translog phase of peer-recovery.
With this change, we guarantee that the value of max seq_no_of_updates
on a replica when any index/delete operation is performed at least the
max_seq_no_of_updates on the primary when that operation was executed.
Relates #33656
If numWrites is between 2 and 9, we will issue an invalid range because
the from_seq_no is negative. This commit makes sure that numWrites is at
least 10, and adds an explicit test to verify invalid request ranges.
This PR is the first step to use seq_no to optimize indexing operations.
The idea is to track the max seq_no of either update or delete ops on a
primary, and transfer this information to replicas, and replicas use it
to optimize indexing plan for index operations (with assigned seq_no).
The max_seq_no_of_updates on primary is initialized once when a primary
finishes its local recovery or peer recovery in relocation or being
promoted. After that, the max_seq_no_of_updates is only advanced internally
inside an engine when processing update or delete operations.
Relates #33656
