Tikv用于juicefs的问题。

TiDBer_MHz90Gih · 2025 年7 月 13 日 04:04

tikv持久化使用的是ceph rbd，在压测juicefs时候，tikv作为meta的lat略高，会影响到读写性能，使用k8s部署了10节点的tikv和3节点的pd，tikv的每个pod的limit在20c、40G内存的情况，还是会出现lat高的问题，版本是7.1

小龙虾爱大龙虾 · 2025 年7 月 13 日 06:35

tikv 对磁盘要求比较高，你磁盘性能如何

TiDBer_MHz90Gih · 2025 年7 月 13 日 09:18

ssd的盘

小龙虾爱大龙虾 · 2025 年7 月 14 日 01:21

你可以发现监控数据，慢是慢到多少

TiDBer_MHz90Gih · 2025 年7 月 14 日 02:16

几十到几百，在JUICEFS没有IO时，是0.X.

小龙虾爱大龙虾 · 2025 年7 月 14 日 02:23

你现在是怀疑 tikv 是慢的地方，所以你要看 tikv 的监控，要是没部署监控，可以参考这个：https://docs.pingcap.com/zh/tidb-in-kubernetes/dev/monitor-a-tidb-cluster/

TiDBer_MHz90Gih · 2025 年7 月 14 日 05:04

kbq exec -n tikv juicefs-tikv-pd-0 – /pd-ctl --pd=“127.0.0.1:2379” store | grep “leader_count”

    "leader_count": 33,
    "leader_count": 0,
    "leader_count": 37,

会出现这个。

TiDBer_MHz90Gih · 2025 年7 月 14 日 05:05

[server]
      grpc-compression-type = "gzip"
      grpc-concurrency = 16
      grpc-raft-conn-num = 16
      grpc-stream-initial-window-size = "16MB"
      grpc-keepalive-time = "3s"
      grpc-keepalive-timeout = "1s"
      grpc-timeout = "5s"
      
      [readpool.storage]
      use-unified-pool = false
      high-concurrency = 16
      normal-concurrency = 12
      low-concurrency = 8
      
      [readpool.coprocessor]
      use-unified-pool = true
      
      [storage]
      scheduler-worker-pool-size = 12
      scheduler-pending-write-threshold = "200MB"
      reserve-space = "10GB"
      enable-async-apply-prewrite = true
      block-cache-size = "18GB"
      
      [raftstore]
      store-pool-size = 6
      apply-pool-size = 6
      
      raft-log-gc-tick-interval = "500ms"
      raft-log-gc-threshold = 32
      raft-log-gc-count-limit = 6000
      raft-log-gc-size-limit = "64MB"
      
      snap-mgr-gc-tick-interval = "30s"
      snap-max-write-bytes-per-sec = "300MB"
      
      raft-base-tick-interval = "500ms"
      raft-heartbeat-ticks = 2
      raft-election-timeout-ticks = 20
      
      raft-max-size-per-msg = "8MB"
      raft-max-inflight-msgs = 256
      raft-entry-max-size = "8MB"
      
      [coprocessor]
      region-max-size = "64MB"
      region-split-size = "48MB"
      region-max-keys = 640000
      region-split-keys = 480000
      
      [pd-client]
      pd-retry-interval = "100ms"
      pd-retry-max-count = 20
      
      [rocksdb]
      max-open-files = 40960
      max-background-jobs = 12
      max-background-flushes = 6
      max-sub-compactions = 6
      max-manifest-file-size = "128MB"
      create-if-missing = true
      
      db-write-buffer-size = "512MB"
      write-buffer-size = "128MB"
      max-write-buffer-number = 6
      min-write-buffer-number-to-merge = 2
      
      compression-per-level = ["no", "no", "lz4", "lz4", "lz4", "zstd", "zstd"]
      bottommost-level-compression = "zstd"
      
      [rocksdb.defaultcf]
      compression-per-level = ["no", "no", "lz4", "lz4", "lz4", "zstd", "zstd"]
      write-buffer-size = "128MB"
      max-write-buffer-number = 6
      min-write-buffer-number-to-merge = 2
      
      level0-file-num-compaction-trigger = 2
      level0-slowdown-writes-trigger = 16
      level0-stop-writes-trigger = 20
      max-bytes-for-level-base = "256MB"
      max-bytes-for-level-multiplier = 6
      target-file-size-base = "64MB"
      target-file-size-multiplier = 1
      
      block-size = "16KB"
      block-cache-size = "15GB"
      cache-index-and-filter-blocks = true
      pin-l0-filter-and-index-blocks = true
      
      bloom-filter-bits-per-key = 10
      whole-key-filtering = true
      
      [rocksdb.writecf]
      compression-per-level = ["no", "no", "lz4", "lz4", "lz4", "zstd", "zstd"]
      write-buffer-size = "256MB"
      max-write-buffer-number = 6
      min-write-buffer-number-to-merge = 2
      level0-file-num-compaction-trigger = 2
      level0-slowdown-writes-trigger = 16
      level0-stop-writes-trigger = 20
      target-file-size-base = "32MB"
      max-bytes-for-level-base = "128MB"
      block-cache-size = "3GB"
      
      [rocksdb.lockcf]
      write-buffer-size = "32MB"
      max-write-buffer-number = 4
      min-write-buffer-number-to-merge = 2
      block-cache-size = "512MB"
      
      [titan]
      enabled = false