DynamoDB call not record after enabling graphql dataloader

[Blankll]

DynamoDB call not record after enabling graphql dataloader

[bizob2828]

@Blankll I responded to all the issues you logged except for this. Can you update your example app to use dataloader so I can check out what's going on around losing context with dynamo?

[workato-integration]

Jira CommentId: 138941
Commented by revans:

It looks like our sync doesn't load all comments. this one is the most relevant thus far #215 (comment)

[jordigh]

Hi, @Blankll thanks for the bug report. This seems related to other problems, and it would help if we can gather more information about what is happening.

For issues like this and other related issues, (newrelic/node-newrelic-aws-sdk#145, newrelic/node-newrelic-aws-sdk#145 newrelic/node-newrelic-aws-sdk#146 and newrelic/node-newrelic-aws-sdk#147), I recommend leveraging our support team over at support.newrelic.com which can help deep-dive into solving the problem. They will also ask to collect the necessary information which may or may not be sensitive (such as logs).

[Blankll]

Hi @jordigh the other 3 issues are more about missing some information in Span, but this one is different, I think this one is because this plugin currently hasn't compatible with dataloader or dataloader should also need to instrument, it is not some information missing in Span, it does not record the DynamoDB call Span that triggered in dataloader at all

[bizob2828]

@Blankll if you can provide reproduction cases that would be helpful. If you're not comfortable doing that, as @jordigh suggested you should go through support.newrelic.com and they can help deep dive.

[Blankll]

Hi @jordigh @bizob2828 this repo has the sample to simulate issue #145, issue #146, and issue #147, please have a check

[bizob2828]

Much appreciated @Blankll. We will take a look internally and get back to you.

[olivierlacan]

I don't think this issue is specific to DynamoDB, I believe it's specific to Dataloader itself and the way it executes all datastore queries as a batch: https://github.com/graphql/dataloader#batching

With this plugin enabled, New Relic database traces misreport time spent loading data within dataloader batch loading functions by a fairly wide margin:

I can confirm the 762ms total load time is accurate based on browser timings of this same GraphQL resolver query:

This also fails to list a Postgres select within each of the resolve.ApolloServer function calls which each correspond to a batch loader function in a Dataloader class.

Since Dataloader uses process.nextTick this plugin (or the newrelic package, maybe?) might need to account for that. See: https://github.com/graphql/dataloader/blob/588a8b6c6391aad042b369f10dc440c7e0458312/src/index.js#L232-L246

[bizob2828]

@olivierlacan if you can create a sample app that reproduces this problem we'd be able to triage the issue much easier. Do you think that's possible? I suspect it's an issue with async context propagation and perhaps adding new behavior for this dataloader

[workato-integration]

https://issues.newrelic.com/browse/NEWRELIC-6377

[bizob2828]

I removed the bug flag as we technically don't support the graphql dataloader plugin

[workato-integration]

Jira CommentId: 199168
Commented by jgutierrezhermoso:

I'm unable to reproduce this issue. I built a sample app using Apollo, Koa, the Dataloader and DynamoDB, and as far as I can tell, everything is working fine. Distributed traces are showing up with DynamoDB in them.

newrelic/newrelic-node-examples#68 (comment)