Was this page helpful?
You're viewing documentation for a previous version of Scylla Manager. Switch to the latest stable version.
This document contains examples for scheduling and running repairs.
A cluster, added to Scylla Manager, has a repair task created for it. This task repairs the entire cluster every week. You can change it, add additional repairs, or delete it. You can schedule repairs to run in the future on a regular basis or schedule repairs to run once as you need them. Any repair can be rescheduled, paused, resumed, or deleted.
While the most recommended way to run a repair is across an entire cluster, repairs can be scheduled to run on a single/multiple datacenters, keyspaces, or tables. The selection mechanism, based on the glob patterns, gives you a lot of flexibility. Scheduled repairs run every X days depending on the frequency you set. Additional parameters are described in the sctool repair command reference.
Use the example below to run the sctool repair command.
sctool repair -c <id|name> [-s <date>] [-i <time-unit>]
-c - the name you used when you created the cluster
-s - the time you want the repair to begin
-i - the time interval you want to use in between consecutive repairs
The command returns the task ID. You will need this ID for additional actions.
If you want to run the repair only once, leave out the interval argument (
In case when you want the repair to start immediately, but you want it to schedule it to repeat at a determined interval, leave out the start flag (
-s) and set the interval flag (
-i) to the time you want the repair to reoccur.
This command will schedule a repair in 4 hours, repair will be repeated every week.
sctool repair -c prod-cluster -s now+4h -i 7d
Command returns the task ID (repair/41014974-f7a7-4d67-b75c-44f59cdc5105, in this case). This ID can be used to query the status of the repair task, to defer the task to another time, or to cancel the task See Managing Tasks.
In order to schedule repair of particular data center, you have to specify
You can specify more than one DC, or use glob pattern to match multiple DCs or exclude some of them.
For Example, you have the following DCs in your cluster: dc1, dc2, dc3
In this example, only dc1 is repaired. The repair repeats every 5 days.
sctool repair -c prod-cluster -i 5d --dc 'dc1'
sctool repair -c prod-cluster -i 5d --dc '*,!dc2'
In order to schedule repair of particular keyspace or table, you have to provide
You can specify more than one keyspace/table or use glob pattern to match multiple keyspaces/tables or exclude them.
sctool repair -c prod-cluster -K 'auth_service.*,!auth_service.lru_cache' --dc 'dc1'
In this example, you repair only token ranges replicated by the node with IP
sctool repair -c prod-cluster --host 126.96.36.199
An ad-hoc repair runs immediately and does not repeat. This procedure shows the most frequently used repair commands. Additional parameters can be used. Refer to repair parameters.
To run an immediate repair on the prod-cluster cluster, saving the repair in my-repairs, run the following command
-c cluster flag with your cluster’s cluster name or ID and replace the
-L flag with your repair’s location:
sctool repair -c prod-cluster -L 's3:my-repairs'
We recommend to use
--dry-run parameter prior scheduling a repair if you specify datacenter, keyspace or table filters.
It’s a useful way to verify that all the data you want will be repaired.
Add the parameter to the end of your repair command, so if it works, you can erase it and schedule the repair with no need to make any other changes.
If you do tables filtering you can pass
--show-tables flag in order to print the table names next to keyspaces.
If the dry run completes successfully, a summary of the repair is displayed. For example:
sctool repair -c prod-cluster -K system*,test_keyspace.* --dry-run
NOTICE: dry run mode, repair is not scheduled
- system_auth (3 tables)
- system_distributed (3 tables)
- system_traces (5 tables)
- test_keyspace (10 tables)
Note that if a keyspace has no tables or a table is empty it will not be listed here. Nevertheless you can still schedule the repair, the glob patterns are evaluated before each repair run so when data is there it will be repaired.
Progress of the repair task can be monitored by using the sctool progress command and providing UUID of the repair task.
sctool progress repair/143d160f-e53c-4890-a9e7-149561376cfd -c prod-cluster
On this page