Improve safety for weed shells `ec.encode`.
The current process for `ec.encode` is:
1. EC shards for a volume are generated and added to a single server
2. The original volume is deleted
3. EC shards get re-balanced across the entire topology
It is then possible to lose data between #2 and #3, if the underlying volume storage/server/rack/DC
happens to fail, for whatever reason. As a fix, this MR reworks `ec.encode` so:
* Newly created EC shards are spread across all locations for the source volume.
* Source volumes are deleted only after EC shards are converted and balanced.
Improve parallelization for `ec.encode`.
Instead of processing one volume at at time, perform all EC conversion
steps (mark readonly -> generate EC shards -> delete volume -> remount) in
parallel for all of them.
This should substantially improve performance when EC encoding
entire collections.
Effectively undoes c9399a68; with ff8bd862, a replica placement type `000`
will no longer break shards re-balancing.
Co-authored-by: Chris Lu <chrislusf@users.noreply.github.com>
The replica placement type specifies numebr of _replicas_ on the same/different rack;
that means we can have one EC shard copy on each, even if the replica setting is zero.
This PR reworks replica placement parsing for EC rebalancing, so we check allow
(replica placement + 1) when selecting racks and nodes to balance EC shards into.
This logic was originally part of `spreadEcShards()`, which got removed during
the unification effort with `ec.balance` (https://github.com/seaweedfs/seaweedfs/pull/6344),
accidentally breaking functionality in the process.
The commit restores the deletion code for EC'd volumes - with parallelization support.
Rework `shell.EcBalance()`'s waitgroup with errors code into a standalone type.
We'll re-use this for other EC jobs - for example, volume creation. Also fixes
potential concurrency issues when collecting error results.
* Begin implementing EC balancing parallelization support.
Impacts both `ec.encode` and `ec.balance`,
* Nit: improve type naming.
* Make the goroutine workgroup handler for `EcBalance()` a bit smarter/error-proof.
* Nit: unify naming for `ecBalancer` wait group methods with the rest of the module.
* Fix concurrency bug.
* Fix whitespace after Gitlab automerge.
* Delete stray TODO.
* Account for replication placement settings when balancing EC shards within racks.
* Update help contents for `ec.balance`.
* Add a few more representative test cases for `pickEcNodeToBalanceShardsInto()`.
* Improve EC shards rebalancing logic across nodes.
- Favor target nodes with less preexisting shards, to ensure a fair distribution.
- Randomize selection when multiple possible target nodes are available.
- Add logic to account for replication settings when selecting target nodes (currently disabled).
* Fix minor test typo.
* Clarify internal error messages for `pickEcNodeToBalanceShardsInto()`.
Improve EC shards rebalancing logic across racks.
- Favor target shards with less preexisting shards, to ensure a fair distribution.
- Randomize selection when multiple possible target shards are available.
- Add logic to account for replication settings when selecting target shards (currently disabled).
* Rename `command_ec_encode_test.go` to `command_ec_common_test.go`.
All tests defined in this file are now for `command_ec_common.go`.
* Minor code cleanups.
- Fix broken `ec.balance` test.
- Rework integer ceiling division to not use floats, which can introduce precision errors.
* Introduce logic to resolve volume replica placement within EC rebalancing.
This will be used to make rebalancing logic topology-aware.
* Give shell.EcNode.dc a dedicated DataCenterId type.
* Refactor `ec.balance` logic into a `weeed/shell/command_ec_common.go` standalone function.
This is a prerequisite to unify the balance logic for `ec.balance` and `ec.encode'.
* s/Balance()/EcBalance()/g
Lisandro Pin