Don't log mass build output

[Nemo157]

When building some crates they cause hundreds/thousands of lines of output from rustc/rustdoc which overwhelms journald and it starts dropping messages. One example is when building pallas@0.13.4 for the non-x86-64 targets:

Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209405Z  INFO log: [stderr] 89  |             d = _mm256_permute4x64_epi64(d, _MM_SHUFFLE(1, 0, 3, 2)); log.target="rustwid
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209424Z  INFO log: [stderr]     |                 ^^^^^^^^^^^^^^^^^^^^^^^^ not found in this scope log.target="rustwide::cm
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209444Z  INFO log: [stderr] ... log.target="rustwide::cmd" log.module_path="rustwide::cmd" log.file="/home/ubuntu/.cargo/re
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209464Z  INFO log: [stderr] 312 |     ROUND!(1, load1!()); log.target="rustwide::cmd" log.module_path="rustwide::cmd" log.f
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209486Z  INFO log: [stderr]     |     ------------------- in this macro invocation log.target="rustwide::cmd" log.module_pa
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209506Z  INFO log: [stderr]     | log.target="rustwide::cmd" log.module_path="rustwide::cmd" log.file="/home/ubuntu/.cargo/
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209524Z  INFO log: [stderr]     = note: this error originates in the macro `UNDIAGONALIZE` which comes from the expansion o
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209545Z  INFO log: [stderr] help: consider importing this function log.target="rustwide::cmd" log.module_path="rustwide::cm
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209564Z  INFO log: [stderr]     | log.target="rustwide::cmd" log.module_path="rustwide::cmd" log.file="/home/ubuntu/.cargo/
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209582Z  INFO log: [stderr] 1   | use core::arch::x86::_mm256_permute4x64_epi64; log.target="rustwide::cmd" log.module_path
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209604Z  INFO log: [stderr]     | log.target="rustwide::cmd" log.module_path="rustwide::cmd" log.file="/home/ubuntu/.cargo/
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.209624Z  INFO log: [stderr]  log.target="rustwide::cmd" log.module_path="rustwide::cmd" log.file="/home/ubuntu/.cargo/regis
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.213346Z  INFO log: [stderr] error[E0425]: cannot find function `_mm256_permute4x64_epi64` in this scope log.target="rustwid
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.213430Z  INFO log: [stderr]    --> /opt/rustwide/cargo-home/registry/src/github.com-1ecc6299db9ec823/cryptoxide-0.4.2/src/h
Nov 14 20:57:58 docsrs cratesfyi[10930]: 2022-11-14T20:57:58.213458Z  INFO log: [stderr]     | log.target="rustwide::cmd" log.module_path="rustwide::cmd" log.file="/home/ubuntu/.cargo/
Nov 14 20:58:20 docsrs cratesfyi[10930]: 2022-11-14T20:58:20.687122Z DEBUG docs_rs::utils::daemon: Checking new crates
Nov 14 20:58:20 docsrs systemd-journald[27648]: Suppressed 7196 messages from docs.rs.service

I would just say we shouldn't log these at all since the build-log stored in S3 will capture them, but I have found it useful to see the output in the logs to diagnose issues on non-default targets while we don't have #787. A halfway point until that is done might be to only log the first 50 lines of build output if possible, that seems like enough to generally diagnose basic issues.

[syphar]

debugging directly on the build-server will probably become less and less, especially when we have #787 ( and #1011 where we probably would also log metadata errors).

In the final picture IMO journald (and after we have multiple build-servers, the log aggregation that we probably have) should only log things that are not in the build logs on S3.

A halfway point until that is done might be to only log the first 50 lines of build output if possible, that seems like enough to generally diagnose basic issues.

this sounds like a good compromise

[syphar]

@Nemo157 just a thought: this is probably mostly a rustwide change, which we then would just use here, right?

[syphar]

Ah no, the output is coming from our config of the log crate, so that's what we have to tweak.

[syphar]

With #3203 we allowed disabling the build logs in the stdout stream,

and disabled it on prod