misc: Propagate full IO stats (count/min/max) to runtime metrics

[rui-mo]

This PR enhances the propagation of IO statistics into RuntimeMetric by
including not just the sum, but also count, min, and max values from the IO
layer. This affects the IO-related metrics such as:prefetchBytes,
ioWaitWallNanos, storageReadBytes, localReadBytes, ramReadBytes.
Previously, these Runtime metrics had a default count of 1, and min/max were
equal to sum because no detailed IO stats were passed down. After this change,
count/min/max now reflect real observed values from the IO layer.

The metrics maxSingleIoWaitWallNanos and numStorageRead are removed
because their functionality is represented by the ioWaitWallNanos.max and
storageReadBytes.count respectively.

Users relying on these metrics for historical analysis, monitoring, or
dashboards should be aware of this impact.

[netlify]

✅ Deploy Preview for meta-velox canceled.

Name	Link
🔨 Latest commit	1ec718b
🔍 Latest deploy log	https://app.netlify.com/projects/meta-velox/deploys/690c8d1d36ed310008692843

[jinchengchenghh]

Could update the PR description?

[Yuhta]

The change itself makes sense, the only caveat is it changes the semantics of these stats drastically (so that all the stats other than sum are not comparable before & after this change). This is something worthing noting in the release note.

[meta-codesync]

@peterenescu has imported this pull request. If you are a Meta employee, you can view this in D86311108.

[Yuhta]

@rui-mo The test failures are relevant, can you double check and fix?

[peterenescu]

To be specific, there are four velox exec tests failing:
PrintPlanWithStatsTest (innerJoinWithTableScan, partialAggregateWithTableScan, tableWriterWithTableScan)
TableScanTest (statsBasedFilterReorderDisabled)

[FelixYBW]

With this change, we may remove maxSingleIoWaitWallNanos, numStorageRead. They are shown in the ioWaitWallNanos.max and storageReadBytes.count

[rui-mo]

Hi everyone, thanks for the suggestions! I've updated the PR description and resolved the unit test failure. Could you please review it again?

[meta-codesync]

@peterenescu merged this pull request in a0d77b3.

[majetideepak]

This PR is causing the values to overflow since IoCounter uses uint64_t and RuntimeMetric uses int64_t.
@rui-mo will you be able to help with a fix?

Specifically due to min_ default in the IoCounter.

std::atomic<uint64_t> min_{std::numeric_limits<uint64_t>::max()};

"TableScan.0.ioWaitWallNanos": {
                     "name": "TableScan.0.ioWaitWallNanos",
                     "unit": "NANO",
                     "sum": 0,
                     "count": 0,
                     "max": 0,
                     "min": -1000
                 }

See prestodb/presto#26588 (comment)

[rui-mo]

@majetideepak Thanks for pointing this out. I opened a PR that changes RuntimeMetrics counters from signed to unsigned (int64_t -> uint64_t) #15536. Let me know your thoughts, thanks.