show when a repair is paused

[amnonh]

It could be useful to see if a repair process is paused.

[tzach]

@amnonh

1. do we have a metric to support this?
2. What is the best way to present this?
   For Enterprise (but not for cloud), there could be more than one repair task in parallel.

[Michal-Leszczynski]

We could add SM metric like scylla_manager_scheduler_status labeled by cluster_id, task_type, task_id which would store enum of all possible task statuses:

// Status enumeration.
const (
	StatusNew      Status = "NEW" // 1
	StatusRunning  Status = "RUNNING"
	StatusStopping Status = "STOPPING"
	StatusStopped  Status = "STOPPED"
	StatusWaiting  Status = "WAITING"
	StatusDone     Status = "DONE"
	StatusError    Status = "ERROR"
	StatusAborted  Status = "ABORTED" // 8
)

[amnonh]

I have concerns about the explosion of metrics from the manager when we have lots of tables.
It's not helpful for the users; we cannot visualize thousands of metrics. We need to have an aggregated version.

[Michal-Leszczynski]

Proposed metric wouldn't be labeled by host or table, so it shouldn't generate much entries.

[tarzanek]

some clusters can have 1200 tables

[ruthea]

some clusters can have 1200 tables

Does this request need to factor in tables? Or is this simply a boolean on whether there is a pause on repair for a cluster?

[tzach]

@tarzanek, does anyone you know care about the table-by-table repair?
Scylla Manager default is to repair all tables
It is possible to change, but I'm not sure anyone does that, for sure not in Scylla Cloud.

[tarzanek]

Usually people want whole cluster stats, so @tzach is right
per table repair is rare activity

[tzach]

Thanks @tarzanek

Look like we have 3 separate issues:

• Enable cluster level metrics
• Disable Tablet level metrics
• Show paused repairs

[amnonh]

@Michal-Leszczynski how are we with regards to those metrics? may be related to #2307

[Michal-Leszczynski]

I suppose that we can put scylladb/scylla-manager#3744 in 3.3.1 milestone.
@karol-kokoszka what do you think?

[karol-kokoszka]

Let's bring this issue to the refirement.
We haven't define any ETA for 3.3.1, so we can put whatever we think is valid to fix.