TiDB Self-Managed
Learning Center
Contact Us
Sign InStart for Free

PITR Monitoring and Alert

PITR supports using Prometheus to collect monitoring metrics. Currently all monitoring metrics are built into TiKV.

Monitoring configuration

  • For clusters deployed using TiUP, Prometheus automatically collects monitoring metrics.
  • For clusters deployed manually, follow the instructions in TiDB Cluster Monitoring Deployment to add TiKV-related jobs to the scrape_configs section of the Prometheus configuration file.

Monitoring metrics

MetricsTypeDescription
tikv_log_backup_interal_actor_acting_duration_secHistogramThe duration of handling all internal messages and events.
message :: TaskType
tikv_log_backup_initial_scan_reasonCounterStatistics of the reasons why initial scan is triggered. The main reason is leader transfer or Region version change.
reason :: {"leader-changed", "region-changed", "retry"}
tikv_log_backup_event_handle_duration_secHistogramThe duration of handling KV events. Compared with tikv_log_backup_on_event_duration_seconds, this metric also includes the duration of internal conversion.
stage :: {"to_stream_event", "save_to_temp_file"}
tikv_log_backup_handle_kv_batchHistogramRegion-level statistics of the sizes of KV pair batches sent by Raftstore.
tikv_log_backup_initial_scan_disk_readCounterThe size of data read from the disk during initial scan. In Linux, this information is from procfs, which is the size of data actually read from the block device. The configuration item initial-scan-rate-limit applies to this metric.
tikv_log_backup_incremental_scan_bytesHistogramThe size of KV pairs actually generated during initial scan. Because of compression and read amplification, this value might be different from that of tikv_log_backup_initial_scan_disk_read.
tikv_log_backup_skip_kv_countCounterThe number of Raft events being skipped during the log backup because they are not helpful to the backup.
tikv_log_backup_errorsCounterThe errors that can be retried or ignored during the log backup.
type :: ErrorType
tikv_log_backup_fatal_errorsCounterThe errors that cannot be retried or ignored during the log backup. When an error of this type occurs, the log backup is paused.
type :: ErrorType
tikv_log_backup_heap_memoryGaugeThe memory occupied by events that are unconsumed and found by initial scan during log backup.
tikv_log_backup_on_event_duration_secondsHistogramThe duration of storing KV events to temporary files.
stage :: {"write_to_tempfile", "syscall_write"}
tikv_log_backup_store_checkpoint_tsGaugeThe store-level checkpoint TS, which is deprecated. It is close to the GC safepoint registered by the current store.
task :: string
tidb_log_backup_last_checkpointGaugeThe global checkpoint TS. It is a time point till which log data has been backed up.
task :: string
tikv_log_backup_flush_duration_secHistogramThe duration of moving local temporary files to the external storage.
stage :: {"generate_metadata", "save_files", "clear_temp_files"}
tikv_log_backup_flush_file_sizeHistogramStatistics of the sizes of files generated during the backup.
tikv_log_backup_initial_scan_duration_secHistogramThe statistics of the overall duration of initial scan.
tikv_log_backup_skip_retry_observeCounterStatistics of the errors that can be ignored during log backup, or the reasons why retry is skipped.
reason :: {"region-absent", "not-leader", "stale-command"}
tikv_log_backup_initial_scan_operationsCounterStatistics of RocksDB-related operations during initial scan.
cf :: {"default", "write", "lock"}, op :: RocksDBOP
tikv_log_backup_enabledCounterWhether to enable log backup. If the value is greater than 0, log backup is enabled.
tikv_log_backup_observed_regionGaugeThe number of Regions being listened to.
tikv_log_backup_task_statusGaugeThe status of the log backup task. 0 means running. 1 means paused. 2 means error.
task :: string
tikv_log_backup_pending_initial_scanGaugeStatistics of pending initial scans.
stage :: {"queuing", "executing"}

Grafana configuration

  • For clusters deployed using TiUP, the Grafana dashboard contains the PITR panel. The Backup Log panel in the TiKV-Details dashboard is the PITR panel.
  • For clusters deployed manually, refer to Import a Grafana dashboard and upload the tikv_details JSON file to Grafana. Then find the Backup Log panel in the TiKV-Details dashboard.

Alert configuration

Currently, PITR does not have built-in alert items. This section introduces how to configure alert items in PITR and recommends some items.

To configure alert items in PITR, follow these steps:

  1. Create a configuration file (for example, pitr.rules.yml) for the alert rules on the node where Prometheus is located. In the file, fill in the alert rules according to the Prometheus documentation, the following recommended alert items, and the configuration sample.
  2. In the rule_files field of the Prometheus configuration file, add the path of the alert rule file.
  3. Send SIGHUP signal to the Prometheus process (kill -HUP pid) or send an HTTP POST request to http://prometheus-addr/-/reload (before you send the HTTP request, add the --web.enable-lifecycle parameter when starting Prometheus).

The recommended alert items are as follows:

LogBackupRunningRPOMoreThan10m

  • Alert item: max(time() - tidb_log_backup_last_checkpoint / 262144000) by (task) / 60 > 10 and max(tidb_log_backup_last_checkpoint) by (task) > 0 and max(tikv_log_backup_task_status) by (task) == 0
  • Alert level: warning
  • Description: The log data is not persisted to the storage for more than 10 minutes. This alert item is a reminder. In most cases, it does not affect log backup.

A configuration sample of this alert item is as follows:

groups:
- name: PiTR
  rules:
  - alert: LogBackupRunningRPOMoreThan10m
    expr: max(time() - tidb_log_backup_last_checkpoint / 262144000) by (task) / 60 > 10 and max(tidb_log_backup_last_checkpoint) by (task) > 0 and max(tikv_log_backup_task_status) by (task) == 0
    labels:
      severity: warning
    annotations:
      summary: RPO of log backup is high
      message: RPO of the log backup task {{ $labels.task }} is more than 10m

LogBackupRunningRPOMoreThan30m

  • Alert item: max(time() - tidb_log_backup_last_checkpoint / 262144000) by (task) / 60 > 30 and max(tidb_log_backup_last_checkpoint) by (task) > 0 and max(tikv_log_backup_task_status) by (task) == 0
  • Alert level: critical
  • Description: The log data is not persisted to the storage for more than 30 minutes. This alert often indicates anomalies. You can check the TiKV logs to find the cause.

LogBackupPausingMoreThan2h

  • Alert item: max(time() - tidb_log_backup_last_checkpoint / 262144000) by (task) / 3600 > 2 and max(tidb_log_backup_last_checkpoint) by (task) > 0 and max(tikv_log_backup_task_status) by (task) == 1
  • Alert level: warning
  • Description: The log backup task is paused for more than 2 hours. This alert item is a reminder and you are expected to run br log resume as soon as possible.

LogBackupPausingMoreThan12h

  • Alert item: max(time() - tidb_log_backup_last_checkpoint / 262144000) by (task) / 3600 > 12 and max(tidb_log_backup_last_checkpoint) by (task) > 0 and max(tikv_log_backup_task_status) by (task) == 1
  • Alert level: critical
  • Description: The log backup task is paused for more than 12 hours. You are expected to run br log resume as soon as possible to resume the task. Log tasks paused for too long have the risk of data loss.

LogBackupFailed

  • Alert item: max(tikv_log_backup_task_status) by (task) == 2 and max(tidb_log_backup_last_checkpoint) by (task) > 0
  • Alert level: critical
  • Description: The log backup task fails. You need to run br log status to see the failure reason. If necessary, you need to further check the TiKV logs.

LogBackupGCSafePointExceedsCheckpoint

  • Alert item: min(tidb_log_backup_last_checkpoint) by (instance) - max(tikv_gcworker_autogc_safe_point) by (instance) < 0
  • Alert level: critical
  • Description: Some data has been garbage-collected before the backup. This means that some data has been lost and is very likely to affect your services.
Download PDF