channelz support might have introduced memory leak

[sasaxing]

Problem description

Upgrading @grpc-js from 1.3.8 to 1.4.1 introduces memory leak.

The image shows the heap space size used (old space)

• green line is using @grpc-js@1.3.8;
• yellow one is @grpc-js@1.4.1.

Reproduction steps

• Set up the most common grpc transport following the guide of nestjs/microservices/grpc (https://docs.nestjs.com/microservices/grpc#call-stream-handler)
• Observe the Used Heap Space Size (Old space) for a few hours (in my case it's leaking ~15M/day)

Environment

• OS name, version and architecture: [docker container x86_64 GNU/Linux Ubuntu 20.04.3 LTS]
• Node version [v16.12.0]
• Package name and version: @grpc-js@1.4.1, consumed by @nestjs/microservice@8.0.11 (same case for 8.1.2)

Additional context

This is the screenshot of the snapshot diff with 15 hours gap in between.
As the object stack indicates, the timestamp pushed into events array each time addTrace is called, it keeps adding new trace but is never garbage collected somehow.

According to the grpc version diff, the problem might be introduced by this commit: grpc-js: Add channelz support, when introducing the new object ChannelzTrace in server.ts after release of 1.4.0.

Is it something now should be disabled explicitly if not needed?

[murgatroid99]

Thank you for the report. This was also reported in #1941, and as a result, I published grpc-js version 1.4.2 with hard limits on the number of retained channelz trace events per channel/subchannel/server, and support for the channel/server option grpc.enable_channelz, which can be set to 0 to disable channelz entirely. Please try it out to see if it solves your problem.

[sasaxing]

looks better. Thanks for the fast fix :)