重新分配达到重试上限的分片（reroute retry_failed）

命令

# Attempt count and last failure details:
GET _cluster/allocation/explain?filter_path=index,shard,unassigned_info.reason,unassigned_info.failed_allocation_attempts,unassigned_info.details,allocate_explanation
# After fixing the cause, retry shards that hit the limit:
POST _cluster/reroute?retry_failed=true&metric=none
# Recovery progress:
GET _cat/recovery?v&active_only=true&h=index,shard,stage,source_node,target_node,bytes_percent,translog_ops_percent

作用

显示因连续失败而停止分配（ALLOCATION_FAILED、max_retry）的分片的失败详情，并再次触发分配；监控恢复进度。

适用场景

allocation explain 显示 'shard has exceeded the maximum number of retries' 时；在磁盘已满、translog 损坏或临时网络问题解决之后。

注意

不修复原因就运行 retry_failed，分片会再尝试 5 次并再次失败；先解决 unassigned_info.details 中的错误。若大量分片同时进入恢复，网络和磁盘负载会上升；受 cluster.routing.allocation.node_concurrent_recoveries 和 indices.recovery.max_bytes_per_sec 限制。