Add uptime metric to TIDB status API

[jomenxiao]

How to know TiDB restarted?(startTS or Uptime)
How to get all of TiDB's status?

[jomenxiao]

#4989

[jackysp]

Good idea, @jomenxiao . We'll consider your advice.

[morgo]

See also: #8842 (mysql compatible way)

[ghost]

Uptime information for components is available in Prometheus. The status port does not explicitly have an uptime metric, but it does expose a start_timestamp at /info.

I am going to leave this request open for an uptime metric. It may be useful to have it pre-computed.

[dveeden]

$ etcdctl get --print-value-only /topology/tidb/127.0.0.1:4000/info | jq '.start_timestamp'
1636957939

$ curl -s http://127.0.0.1:10080/info | jq '.start_timestamp'
1636957939

mysql> SELECT START_TIME, UPTIME FROM information_schema.CLUSTER_INFO WHERE INSTANCE='127.0.0.1:4000';
+---------------------------+------------------+
| START_TIME                | UPTIME           |
+---------------------------+------------------+
| 2021-11-15T07:32:19+01:00 | 55m15.629421432s |
+---------------------------+------------------+
1 row in set (0.01 sec)

(see also #29771 )

However there is a slight issue with this start_timestamp: it is coming from etcd, which doesn't work when running in unistore instead of TiKV/PD. In that case getServerInfo will use time.Now().Unix() instead, so it always shows an uptime of less than a second.

Is the information_schema.CLUSTER_INFO and per-instance start_timestamp from the HTTP API sufficient? Both the information_schema and etcd query will give information about all instances.