*: use address instead of storeID for metrics

[rleungx]

What problem does this PR solve?

Closes #1427.

What is changed and how it works?

This PR is going to use TiKV address instead of storeID for the metrics.

Check List

Tests

• Unit test

Related changes

• Need to update the tidb-ansible repository

[codecov-io]

Codecov Report

Merging #1429 into master will decrease coverage by 0.1%.
The diff coverage is 79.56%.

@@            Coverage Diff             @@
##           master    #1429      +/-   ##
==========================================
- Coverage   67.72%   67.62%   -0.11%     
==========================================
  Files         158      158              
  Lines       15146    15165      +19     
==========================================
- Hits        10258    10255       -3     
- Misses       3965     3980      +15     
- Partials      923      930       +7

Impacted Files	Coverage Δ
server/metrics.go	100% <ø> (ø)	⬆️
server/schedule/metrics.go	100% <ø> (ø)	⬆️
server/schedulers/metrics.go	100% <ø> (ø)	⬆️
server/schedulers/balance_leader.go	91.57% <100%> (ø)	⬆️
server/store_statistics.go	84.95% <100%> (ø)	⬆️
server/schedule/filters.go	76.3% <100%> (ø)	⬆️
server/schedulers/balance_region.go	83.49% <100%> (ø)	⬆️
server/server.go	77.61% <100%> (-0.5%)	⬇️
server/coordinator.go	84.67% <57.14%> (ø)	⬆️
server/grpc_service.go	53.84% <63.63%> (-0.15%)	⬇️
... and 8 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5e81548...1d36398. Read the comment docs.

[disksing]

Why does it need to be cherry-picked to release branch?

[disksing]

Do we need to guide users to clean up old metrics data in release note? It may mess up if old and new format mix together.

[nolouch]

Do we need an API to find the store id?

[rleungx]

@disksing The current release-2.1 branch of TiKV uses instance. I think we need to cherry-pick it to release-2.1 branch of PD to keep consistent with TiKV. And for the label name, how about keeping using store rather than address?

[rleungx]

@nolouch Yes, it will be better. Maybe this work can be done by another PR?

[disksing]

@disksing The current release-2.1 branch of TiKV uses instance. I think we need to cherry-pick it to release-2.1 branch of PD to keep consistent with TiKV. And for the label name, how about keeping using store rather than address?

I think use the same label name won't solve the problem I mentioned. Without clean up old data, there will be both label values of store id and store address.

[rleungx]

@disksing Oh, yes. But since we will change the label name and also change the expression of Grafana, the old data won't appear, right?

[disksing]

I don't think so. You may start a cluster to try it. @rleungx

[rleungx]

I have tested it and the old data won't appear.

[siddontang]

@rleungx

seem now if you cherry pick to release, the user must upgrade the ansible too, right?

[rleungx]

@siddontang Yes. So do we need to cherry-pick it to release-2.1?

[siddontang]

@rleungx

I think we don't need to merge to release.

[rleungx]

Ok.

[rleungx]

/rebuild