Add resource stats to task framework

[sruti1312]

Signed-off-by: sruti1312 srutiparthiban@gmail.com

Description

Add resource stats to task framework. The task expose interfaces to allow updating the resource stats.

Sample task response:

"rh6TkaWmTwCp3Tb0zrjHKQ:8120" : {
     "node" : "rh6TkaWmTwCp3Tb0zrjHKQ",
     "id" : 8120,
     "type" : "direct",
     "action" : "indices:data/read/search[phase/query]",
     "start_time_in_millis" : 1637004446880,
     "running_time_in_nanos" : 9010172401,
     "cancellable" : true,
     "parent_task_id" : "rh6TkaWmTwCp3Tb0zrjHKQ:8119",
     "headers" : { },
     "resource_stats" : {
            "total_resource_consumption" : {
              "cpu_time_in_nanos" : 0,
              "memory_in_bytes" : 0
            }
     }
}     

Issues Resolved

This is used for task resource tracking.
#1009
#1329

Check List

• New functionality includes testing.
  • All tests pass
• New functionality has been documented.
  • New functionality has javadoc added
• Commits are signed per the DCO using --signoff

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

[opensearch-ci-bot]

Can one of the admins verify this patch?

[opensearch-ci-bot]

✅   Gradle Check success fa63613384c4b3e2e19f41b58cd76f175396ad62
Log 2334

Reports 2334

[dblock]

Would it simplify things to make ResourceStats a class that is composed of a collection of ResourceStat? From a cursory overview, It feels like it would move serialization into the class along with other helper methods that operate on the collection of stats, simplify usage, and make some of the enums child classes of ResourceStat. It would also enable extending resource stats in the future into something more than just a map of string/value.

[andrross]

Would it simplify things to make ResourceStats a class that is composed of a collection of ResourceStat?

I agree. In general, when you find yourself creating things like Map<String, Map<String, Long>> it's often a sign that you can improve readability and maintainability by creating dedicated classes with more meaningful names and types.

[sruti1312]

@dblock @andrross
TaskInfo.class includes a parser that is used by REST high level client to parse the task response. This is different as compared to other stats objects like NodeStats or JvmStats. The complexity with creating an object like ResourceStats is that we will not be able to use inbuilt object parsers and create custom parsing logic (token by token) if we want to use the above structure. Different approaches that does not need custom parsers are as follows,

1. Using a list - List<ResourceStats>:

ResourceStats: {
     stats_type::String
     resource_metadata::Map<String, Long>
}

The structure would look like this,

"resource_stats" : {
            "stats_type": "total_resource_consumption",
            "resource_metadata": {
              "cpu_time_in_nanos" : 100,
              "memory_in_bytes" : 100
            },  
            "stats_type": "response_resource_consumption",
             "resource_metadata": {
              "cpu_time_in_nanos" : 50,
              "memory_in_bytes" : 50
            },  
            "stats_type": "operator_resource_consumption",
            "resource_metadata": {
              "cpu_time_in_nanos" : 50,
              "memory_in_bytes" : 50
            }
     }

Note: stats_type's value cannot be used as the key instead of resource_metadata as parser does not support dynamic key values. If we need to support this, we will need custom parsing logic.

2. Current approach - Map<String, Map<String, Long>>

"resource_stats" : {
            "total_resource_consumption" : {
              "cpu_time_in_nanos" : 100,
              "memory_in_bytes" : 100
            },  
            "response_resource_consumption" : {
              "cpu_time_in_nanos" : 50,
              "memory_in_bytes" : 50
            },  
            "operator_resource_consumption" : {
              "cpu_time_in_nanos" : 50,
              "memory_in_bytes" : 50
            }
     }

Any thoughts?

[opensearch-ci-bot]

❌   Gradle Check failure 8e9f1bbb2d5b9b8116831fa021c756051cecc170
Log 2465

Reports 2465

[opensearch-ci-bot]

❌   Gradle Check failure da4b58b61e789b2a18474956212ae37610a99eec
Log 2472

Reports 2472

[opensearch-ci-bot]

❌   Gradle Check failure e8ecfd06c0206c9a52e33fe4aa7dd3f718fe6111
Log 2475

Reports 2475

[opensearch-ci-bot]

❌   Gradle Check failure 67d10ce640623c17d9fb5fafd52c47fc2444739e
Log 2477

Reports 2477

[opensearch-ci-bot]

❌   Gradle Check failure 6f2ebcd240d4dbc6cdb530bb33ab9cc8f5799534
Log 2479

Reports 2479

[opensearch-ci-bot]

❌   Gradle Check failure 5da6e8a5983215f35ebe928581363815a6fdc8c0
Log 2481

Reports 2481

[opensearch-ci-bot]

❌   Gradle Check failure 0bbad3d8edb27fa524c3fb9800a8709a21045fe6
Log 2482

Reports 2482

[opensearch-ci-bot]

❌   Gradle Check failure 9cca9ea49d21a82450fd73dc1fb6b025502d4779
Log 2484

Reports 2484

[andrross]

The complexity with creating an object like ResourceStats is that we will not be able to use inbuilt object parsers and create custom parsing logic

Is it a bad thing to write custom parsing logic? The option you presented above as "current approach" is I believe the correct way to structure the JSON. The question here is how to model Java code. The approaches I would consider are:

1. Current approach: Map<String, Map<String, Long>>

2. Map of Resource stats: Map<String, ResourceUsage>

ResourceUsage {
  long cpuTimeInNanos;
  long memoryInBytes;
}

3. Full custom data structure:

ResourceStats {
  ResourceUsage totalResourceConsumption;
  ResourceUsage responseResourceConsumption;
  ResourceUsage operatorResourceConsumption;
}

ResourceUsage {
  long cpuTimeInNanos;
  long memoryInBytes;
}

[opensearch-ci-bot]

❌   Gradle Check failure 78880ef84769b0309430c34d34032f5b3dbf81e2
Log 3713

Reports 3713

[opensearch-ci-bot]

✅   Gradle Check success b64ce26
Log 3715

Reports 3715

[sruti1312]

@dblock and @nknize I have addressed the comments. Can you help with reviewing this PR?

[dblock]

@Bukhtawar you're good with this?

[Bukhtawar]

We will revisit as needed, no immediate blockers for merge

All changes were made.

[dblock]

@sruti1312 Do we want this in 2.0? 2.x? Let's label backports if so.

[sruti1312]

@dblock We will want this in 2.0. This PR has the v2.0.0 tag. Is there another tag that we need to included specifically for backports? If so can you help with adding it as I do not have permissions to add/update labels.

[dblock]

I labeled to backport 2.x and 2.0, we have auto-backport going on.

[reta]

We probably should not create a new class for that: new TaskResourceStats(Collections.singletonMap(TOTAL, getTotalResourceStats()))