Today, we send operations in phase2 of peer recoveries batch by batch
sequentially. Normally that's okay as we should have a fairly small of
operations in phase 2 due to the file-based threshold. However, if
phase1 takes a lot of time and we are actively indexing, then phase2 can
have a lot of operations to replay.
With this change, we will send multiple batches concurrently (defaults
to 1) to reduce the recovery time.
Backport of #58018
Restoring from a snapshot (which is a particular form of recovery) does not currently take recovery throttling into account
(i.e. the `indices.recovery.max_bytes_per_sec` setting). While restores are subject to their own throttling (repository
setting `max_restore_bytes_per_sec`), this repository setting does not allow for values to be configured differently on a
per-node basis. As restores are very similar in nature to peer recoveries (streaming bytes to the node), it makes sense to
configure throttling in a single place.
The `max_restore_bytes_per_sec` setting is also changed to default to unlimited now, whereas previously it was set to
`40mb`, which is the current default of `indices.recovery.max_bytes_per_sec`). This means that no behavioral change
will be observed by clusters where the recovery and restore settings were not adapted.
Relates https://github.com/elastic/elasticsearch/issues/57023
Co-authored-by: James Rodewig <james.rodewig@elastic.co>
The `indices.recovery.max_bytes_per_sec` recovery bandwidth limit can differ
between nodes if it is not set dynamically, but today this is not obvious. This
commit adds a paragraph to its documentation clarifying how to set different
bandwidth limits on each node.
Co-Authored-By: James Rodewig <james.rodewig@elastic.co>
* Clarify that peer recovery settings apply to shard relocation
* Fix awkward wording of 1st sentence
* [DOCS] Remove snapshot recovery reference.
Call out link to [[cat-recovery]].
Separate expert settings.
Today file-chunks are sent sequentially one by one in peer-recovery. This is a
correct choice since the implementation is straightforward and recovery is
network bound in most of the time. However, if the connection is encrypted, we
might not be able to saturate the network pipe because encrypting/decrypting
are cpu bound rather than network-bound.
With this commit, a source node can send multiple (default to 2) file-chunks
without waiting for the acknowledgments from the target.
Below are the benchmark results for PMC and NYC_taxis.
- PMC (20.2 GB)
| Transport | Baseline | chunks=1 | chunks=2 | chunks=3 | chunks=4 |
| ----------| ---------| -------- | -------- | -------- | -------- |
| Plain | 184s | 137s | 106s | 105s | 106s |
| TLS | 346s | 294s | 176s | 153s | 117s |
| Compress | 1556s | 1407s | 1193s | 1183s | 1211s |
- NYC_Taxis (38.6GB)
| Transport | Baseline | chunks=1 | chunks=2 | chunks=3 | chunks=4 |
| ----------| ---------| ---------| ---------| ---------| -------- |
| Plain | 321s | 249s | 191s | * | * |
| TLS | 618s | 539s | 323s | 290s | 213s |
| Compress | 2622s | 2421s | 2018s | 2029s | n/a |
Relates #33844
Today we throttle recoveries only for incoming recoveries. Nodes that have a lot
of primaries can get overloaded due to too many recoveries. To still keep that at bay
we limit the number of threads that are sending files to the target to overcome this problem.
The right solution here is to also throttle the outgoing recoveries that are today unbounded on
the master and don't start the recovery until we have enough resources on both source and target nodes.
The concurrency aspects of the recovery source also added a lot of complexity and additional threadpools
that are hard to configure. This commit removes the concurrent streamns notion completely and sends files
in the thread that drives the recovery simplifying the recovery code considerably.
Outgoing recoveries are not throttled on the master via a allocation decider.
James Rodewig
Nhat Nguyen
Yannick Welsch
James Rodewig
David Turner
alamzeeshan
Simon Willnauer
Clinton Gormley