build: remove typescript 3.6 and 3.7 support - #36329

@IgorMinar can correct me if I'm wrong, but I believe the next release on master is 10.0.0-next.0. Either way, I'm going to split this PR up into two commits and we can drop the second one if it's not the right time for it:

1. updating the dependency versions for angular to 9.1 and typescript to 3.8.3
2. dropping the 3.6 and 3.7 support

@filipesilva - I've encountered frequent failures on the test_aio/aio_preview jobs due to OOM/unexpected crashes. Could TS 3.8 have more memory pressure that's causing our CI to be unstable?

Some example failures:
https://app.circleci.com/pipelines/github/angular/angular/12455/workflows/ec50f5b4-0651-4348-9be0-0240dfcf77dd/jobs/667485
https://app.circleci.com/pipelines/github/angular/angular/12475/workflows/3c3bcacc-8389-4810-a53b-2d5afc7fc2da/jobs/667753
https://app.circleci.com/pipelines/github/angular/angular/12510/workflows/35da99da-f4f1-48af-837f-58e9a886fcf2/jobs/668391

@atscott I think @clydin also noticed something similar, with regards to Build Optimizer. Correct me if I'm wrong though.

The issue I'm currently investigating relates to a runtime error due to incorrect import eliding. It appears that TS 3.8 when used within build optimizer is incorrectly marking certain imports as unreferenced. However, this only appears to occur within the very specific build optimizer usage.
In regards to memory, I haven't noticed any issues.

From the logs, the errors appear to be occurring during the ngcc parallel worker phase (one of them looks to be with a standalone ngcc execution). Do these errors occur with 9.1 when using TS 3.7? It looks like ngcc will limit the number of process workers to 8 but that may be too much for that CI instance (4 GB of memory). CPU counts on CI systems tend to be inaccurate due to the number of underlying virtual CPUs being quite high (16/32/128/etc.) so even though it's a 2 CPU instance the count will most likely not be 2.

@clydin - based on my very unscientific testing in #36380, it does seem like the issue happens with the update to 9.1, not the TS 3.8 upgrade: https://app.circleci.com/pipelines/github/angular/angular/12538/workflows/5f75318c-510a-4de9-aa86-8862a0efcc30/jobs/668882

FYI, @gkalpak has been investigating ngcc worker crashes, which have also been observed on ngcc-validation CI since 9.1. There's FW-2008 to track that work.

From dealing with similar issues within the CLI's build pipeline, the minimum of 8 workers may be too large in relation to the memory usage profile of the ngcc workers.

Hi @filipesilva and @IgorMinar - PTAL. This PR has been rebased with master since FW-2008 was resolved. I was also having issues with the main yarn.lock so @josephperrott recommended updating the keys to v7 in config.yml.

Hi @filipesilva and @IgorMinar - PTAL. This PR has been rebased with master since FW-2008 was resolved. I was also having issues with the main yarn.lock so @josephperrott recommended updating the keys to v7 in config.yml.

Glad that fixed it. Another +1 to remove the fallback cache. Having an incremental node_modules update on yarn.lock changes has broken things a few times now. @alan-agius4 did you have some measurements on fallback cache performance?

@alan-agius4 looked into the Angular CLI fallback cache in angular/angular-cli#17533 and removed it because it wasn't worth it.

In order to avoid potential CI flakes, this should not be merged before #36145 (which in turn is blocked on getting a released version that includes the fix from #36626).

presubmit

This issue has been automatically locked due to inactivity.
Please file a new issue if you are encountering a similar or related problem.

Read more about our automatic conversation locking policy.

_{This action has been performed automatically by a bot.}