fix: node qps sum inconsistent with app qps sum

[hycdong]

What problem does this PR solve?

In shell tools, we can use app_stat -q to show app's qps, and we also can use nodes -q to show nodes' qps. However, the results of two command are different. This pull request fix this bug and add cu in nodes -q command.

Qps counters are like replica*app.pegasus*[email protected], we should use perf-counters-by-prefix to filter app id and partition index, and accumulate them to get qps of specific node.

We record P99 and P999 latency, those counters are like zion*profiler*RPC_RRDB_RRDB_PUT.latency.server and zion*profiler*RPC_RRDB_RRDB_PUT.latency.server.p999, and latency value won't be accumulated.

As a result, I separate qps and latency in two query-commands, and display them in nodes -q.

Command example

nodes -q command example:

>>> nodes -q
[details]
address      status    get_qps  mget_qps  read_cu  put_qps  mput_qps  write_cu  get_p99(ms)  mget_p99(ms)  put_p99(ms)  mput_p99(ms)
ip1:host1    ALIVE        0.00      0.00     0.00  1014.91      0.00   3510.00         0.00          0.00       336.78          0.00
ip2:host2    ALIVE        0.00      0.00     0.00  1004.73      0.00   2675.00         0.00          0.00       344.20          0.00
ip3:host3    ALIVE        0.00      0.00     0.00   992.71      0.00   3815.00         0.00          0.00       231.92          0.00

[summary]
total_node_count    : 3
alive_node_count    : 3
unalive_node_count  : 0

Check List

Tests

• Unit test
• Integration test
• Manual test