Bug: metadata and dimensions not cleared on publish

[perpil]

Bug description

When you invoke publishStoredMetrics, it clears the storedMetrics, but doesn't clear the metadata or dimensions. My understanding is that metadata and dimensions are per request, where defaultDimensions should persist between requests. The python implementation clears metadata and dimensions on publish.

Expected Behavior

Request 1:

metrics.addMetadata('field', 'value');

Log entry for metric has: field: value

Request 2:

metrics.addMetadata('field2', 'value2');

Log entry for metric has: field2: value2

Current Behavior

Request 1:

metrics.addMetadata('field', 'value');

Log entry for metric has: field: value

Request 2:

metrics.addMetadata('field2', 'value2');

Log entry for metric has: field: value and field2: value2

Possible Solution

Invoke clearMetadata() and clearDimensions() as part of publishStoredMetrics()

Steps to Reproduce

See current behavior

Environment

• Powertools version used: 1.2.1
• Packaging format (Layers, npm): npm
• AWS Lambda function runtime: node 16
• Debugging logs: N/A

[dreamorosi]

Hi @perpil, thank you for taking the time to open this issue.

I have looked into it and I was able to reproduce the behavior described, as well as confirm that the Python version handles this same case differently as you have pointed out.

I have created a repo with a full reproduction example, which can be found here: https://github.com/dreamorosi/repro-issue-1115 (link might become dead after this issue is resolved).

The function example is pretty contrived, but helps to demonstrate the issue:

import {
  Metrics,
  logMetrics,
  MetricUnits,
} from "@aws-lambda-powertools/metrics";
import middy from "@middy/core";

const metrics = new Metrics({ namespace: "someName" });
let requestNumber = 1;

export const handler = middy(async () => {
  switch (requestNumber) {
    case 1:
      metrics.addMetadata("field", "value");
      break;
    case 2:
      metrics.addMetadata("field2", "value2");
      break;
    case 3:
      metrics.addDimension("dim1", "value1");
      break;
    case 4:
      metrics.addDimensions({
        dim2: "val2",
      });
    default:
      break;
  }

  metrics.addMetric("successfulBooking", MetricUnits.Count, 1);
  requestNumber++;

  return;
}).use(logMetrics(metrics, { throwOnEmptyMetrics: true }));

In the function above we keep a count of the request number. We then use this count to conditionally add metadata or dimensions to the metric being emitted.

Since we are using the logMetrics middleware, the metrics are automatically published/flushed at the end of each execution, so we would expect to have each metadata field, or dimension, represented at most once in each metric log entry.

Instead, the metadata and dimensions pile up:

I have then tested a similar function implementation, with the Python version of Powertools:

from aws_lambda_powertools import Metrics
from aws_lambda_powertools.metrics import MetricUnit

class RequestNo:
    def __init__(self):
        self.number = 1
    
    def add(self):
        self.number = self.number + 1

metrics = Metrics(namespace="someName")
request_number = RequestNo()

@metrics.log_metrics(raise_on_empty_metrics=True)
def lambda_handler(event, context):
    metrics.add_metric(name="SuccessfulBooking", unit=MetricUnit.Count, value=1)

    if request_number.number == 1:
        metrics.add_metadata(key="field", value="value")
    elif request_number.number == 2:
        metrics.add_metadata(key="field2", value="value2");
    elif request_number.number == 3:
        metrics.add_dimension(name="dim1", value="value1")
    elif request_number.number == 4:
        metrics.add_dimension(name="dim2", value="value2")
    else:
        pass
    
    request_number.add()
    
    return

In this case the dimensions and metadata were cleared out after each request:

This is indeed a bug. We'll try to apply a fix as soon as we have capacity.

If anyone would like to contribute a fix, please leave a comment below expressing interest and discussing a potential fix & scope of the PR, before opening one.

[shdq]

I would like to help with this issue. Could you assign me, please? I understand the problem, but I'm out of context, so I need some time and help to get into it. Thanks.

[dreamorosi]

Hi @shdq, thanks for taking initiative.

I would suggest to move forward as follows:

• On one side, try to set up the development environment as described in the Setup section of the Contributing docs.
• On the other side, try to familiarise yourself with the Metrics codebase and how we run unit tests on it

After that, maybe you could write down in this issue a high level plan of where/what the changes would be and what kind of unit tests you'd want to add to prove that the issue is fixed and avoid future regressions.

We can have a discussion starting from there, and of course if you have any question on any of the above please feel free to reach out, we're happy to help.

PS: we also have a Discord server (invite link) that you can join to chat in case of questions.

[shdq]

Hi there!

I managed to set up the dev environment. Small feedback: npm run setup-local doesn't install if-node-version and ts-md5 packages out of the box, so you can't run tests (npm run lerna-test:unit) until you install them.

So, I read the docs, looked at the code, got familiar with it, and run the tests. And this is what I suggest.

A test for the metadata:

In the tests/unit/Metrics.test.ts file under the Feature: Metadata block of tests, add a new test Publish Stored Metrics should clear metadata, that checks logs twice before publishStoredMetrics() and after for the metadata item existence.

A test for the dimensions:

In the tests/unit/Metrics.test.ts file under the Feature: Dimensions logging block of tests, add a new test Publish Stored Metrics should clear added dimensions, that checks logs for dimensions array length and dimension item value before and after publishStoredMetrics() invocation.

A fix for this is to invoke clearDimensions(); and clearMetadata(); inside publishStoredMetrics() method.

Stored metrics are flushing using this.storedMetrics = {}; inside publishStoredMetrics() right now. And there is a method clearMetrics() which is doing the same inside. So should I use this.dimensions = {}; and this.metadata = {}; for consistency purposes, or change stored metrics to clearMetrics()?

Should I add some more tests for default dimensions, because as I got from the docs, they're persistent across metrics? It seems like this change is not going to affect default dimensions.

Looking for your response.
Best, Sergei

[dreamorosi]

Hi @shdq, thanks for following up on this and for taking the time to flash out the changes, really appreciate it!

The tests for metadata & dimensions look reasonable. Depending on how you'll implement this we might have to do some small tweaks to get full coverage, but we can move forward for now and see once we get there.

A fix for this is to invoke clearDimensions(); and clearMetadata(); inside publishStoredMetrics() method.

Agree with this, I think that's the way to go.

Stored metrics are flushing using this.storedMetrics = {}; inside publishStoredMetrics() right now. And there is a method clearMetrics() which is doing the same inside. So should I use this.dimensions = {}; and this.metadata = {}; for consistency purposes, or change stored metrics to clearMetrics()?

I'd change the current this.storedMetrics = {}; to call this.clearMetrics(), well spotted!

Should I add some more tests for default dimensions, because as I got from the docs, they're persistent across metrics? It seems like this change is not going to affect default dimensions.

I'd be more inclined to keep this PR focused to fix the bug, we can address this in a separate one. Once the bug fix is merged, if you're still interested in working on this we can work on another PR.

From our side, whenever you are ready, you can move forward with the PR. We'll be happy to review it once it's up :)

Thank you again for your time Sergei!

---

About the missing dependency, you're right. One of the PRs currently open is going to fix that once merged. Sorry about that!

[dreamorosi]

The fix has been released in v1.4.0