Files
seaweedfs/weed/storage/erasure_coding/ec_teardown.go
T
Chris Lu 8714f42abf erasure_coding: share the EC shard teardown primitive (#10740)
The unmount+full-teardown of EC shards was duplicated: the plugin-worker EC
task had unmountAndDeleteEcShards and the shell had unmountAndDeleteEcShardsQuiet,
byte-identical apart from a fence parameter and a sentinel error. That
duplication is how the teardown fence semantics drifted between the two paths.

Distribute, mount and verify already live in weed/storage/erasure_coding and are
shared by both callers; move the teardown there too, as UnmountAndDeleteEcShards
plus the shared ErrFullTeardownNotAcked sentinel. Both paths now call the one
function, so the fence semantics cannot diverge again. The shell keeps a thin
type-converting wrapper and aliases the sentinel; behavior is unchanged.
2026-08-13 10:37:25 -07:00

77 lines
3.0 KiB
Go

package erasure_coding
import (
"context"
"errors"
"fmt"
"github.com/seaweedfs/seaweedfs/weed/operation"
"github.com/seaweedfs/seaweedfs/weed/pb"
"github.com/seaweedfs/seaweedfs/weed/pb/volume_server_pb"
"google.golang.org/grpc"
)
// ErrFullTeardownNotAcked marks a reachable server that completed the delete
// RPC but did not report a full teardown (e.g. a pre-upgrade volume server), so
// a stale EC generation may remain on it. Callers distinguish this from an
// unreachable node (which may recover and be re-swept) with errors.Is.
var ErrFullTeardownNotAcked = errors.New("delete did not perform full teardown (pre-upgrade volume server?); a stale EC generation may remain")
// UnmountAndDeleteEcShards unmounts then tears down the named EC shards for a
// volume on one server. Unmount must precede delete (delete requires the shard
// be unmounted); both RPCs are idempotent against missing shards.
//
// encodeTsNs fences both RPCs:
// - 0 selects the server's blanket, generation-independent teardown. This is
// the correct choice for a pre-encode or rollback wipe: it clears same-
// generation shards (a retried encode's prior attempt shares the job's
// generation) and shards whose .vif generation is unreadable (an
// interrupted distribute never landed the sidecar) — both of which a fenced
// teardown preserves. The blanket path aborts rather than clobber a live
// newer mount, and the caller must guarantee no concurrent newer encode of
// this volume (e.g. the admin dedupe key, or an operator lock).
// - a non-zero value fences the teardown to strictly-older generations,
// preserving same-or-newer, generation 0, and an unreadable .vif — for a
// stale-worker cleanup that must never wipe a newer run's live shards.
//
// Returns ErrFullTeardownNotAcked (wrapped, so errors.Is matches) when a
// reachable server does not ack the full teardown.
//
// This is the single teardown primitive shared by the plugin-worker EC task
// and the shell ec.encode pre-cleanup, so the fence semantics cannot drift
// between the two paths.
func UnmountAndDeleteEcShards(
ctx context.Context,
dialOption grpc.DialOption,
server pb.ServerAddress,
collection string,
volumeID uint32,
shardIds []uint32,
encodeTsNs int64,
) error {
return operation.WithVolumeServerClient(false, server, dialOption,
func(client volume_server_pb.VolumeServerClient) error {
if _, err := client.VolumeEcShardsUnmount(ctx, &volume_server_pb.VolumeEcShardsUnmountRequest{
VolumeId: volumeID,
ShardIds: shardIds,
EncodeTsNs: encodeTsNs,
}); err != nil {
return fmt.Errorf("unmount: %w", err)
}
resp, err := client.VolumeEcShardsDelete(ctx, &volume_server_pb.VolumeEcShardsDeleteRequest{
VolumeId: volumeID,
Collection: collection,
ShardIds: shardIds,
FullTeardown: true,
EncodeTsNs: encodeTsNs,
})
if err != nil {
return fmt.Errorf("delete: %w", err)
}
if !resp.GetFullTeardownDone() {
return fmt.Errorf("delete on %s: %w", server, ErrFullTeardownNotAcked)
}
return nil
})
}