apm v3

[vmarchaud]

No description provided.

[vmarchaud]

Breaking changes:

• Removed io.scopedAction
• Removed io.notify in favor of io.notifyError
• Removed support for gc-stats module
• Removed heap snapshot when using the inspector and node 11
• Removed Heap profiling support when using the profiler addon
• Removed deep-metrics support
• Removed Entrypoint class
  - Removed the possibility to set the value of a metrics via a function, now user should set the value when they want to update a metrics.
• Removed io.transpose
• When creating multiple metrics at the same time, metrics will now be returned in an array to allow simpler destruction
• Changed the io.init options to be easier to understand and manage in the code
• Allowed to use the inspector-based in node 8 if running in production (we assume that no other debugger is running in production)

Notable changes:

• Introduced the tracing system without PM2 (still using the old tracing agent)
• Added support for @pm2/node-runtime/stats to collect GC stats, libuv metrics and kernel metrics
• Added error middleware for Koa and fixed the one for Express
• When destroying a pmx instance, actually remove everything listener that we have on all kind of modules so we don't leak something when instancing multiples times
• Added P95 metrics about http latency and event loop ones.
• Added metrics about Heap usage, Heap limit and Heap current.
• Node 11 now support custom actions
• Added a LOT, LOT of logging about what's happening, just use DEBUG=axm:* to toggle it

Improvements;

• Remove easily replaceable dependency
• Use typescript to run the tests (before test were compiled before ran)
• Use Idle notifier of Nodejs to better vizualise iddle time in CPU profiling
• Type completion when using any methods from @pm2/io
• Metrics that are created but never used will not be sended to backend

Notes:

• scopedAction implementation is still there, just not exposed in the API
• Entrypoint was "removed" because i didn't understand the implementation while refactoring, but could be trivially re-implemented

[Unitech]

• event loop latency + event loop latency p95 should be floored at the second decimal
• Camel Case for heap size + used heap size

what about a metric for RSS?

[vmarchaud]

@Unitech RSS is the memory metric in case of using the standalone app, so i don't think it's useful to re-add it as a separate metric