cmd,common,server: Add LP_EXTEND_TIMEOUTS environment variable

[leszko]

What does this pull request do? Explain your changes. (required)

Add env variable LP_EXTEND_TIMEOUTS which extends all broadcaster's timeouts to 8s.

This is a hack to fix Test Streams, so the parameter is not exposed as a flag (or in config file). It's planned to be temporary until the poper solution for Test Streams is implemented: livepeer/stream-tester#145

Specific updates (required)

How did you test each of these updates (required)

Added printlns with timeouts and tested on local devnet

Does this pull request close any open issues?

fix livepeer/stream-tester#144

Checklist:

• Read the contribution guide
• make runs successfully
• All tests in ./test.sh pass
• README and other documentation updated
• Pending changelog updated

[codecov]

Codecov Report

Merging #2392 (3111d01) into master (25c2295) will not change coverage.
The diff coverage is 100.00000%.

@@              Coverage Diff              @@
##              master       #2392   +/-   ##
=============================================
  Coverage   54.94227%   54.94227%           
=============================================
  Files             93          93           
  Lines          19313       19313           
=============================================
  Hits           10611       10611           
  Misses          8113        8113           
  Partials         589         589           

Impacted Files	Coverage Δ
common/util.go	55.70470% <ø> (ø)
server/segment_rpc.go	78.75186% <100.00000%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 25c2295...3111d01. Read the comment docs.

[yondonfu]

This looks like a reasonable change to me, but going to tag some network video folks as well for a second look and for awareness.

[AlexKordic]

LGTM!

[AlexKordic]

@leszko Why is longer timeout needed?

[leszko]

@leszko Why is longer timeout needed?

It's only for the stream tester. Here's the context: livepeer/stream-tester#144

[victorges]

LGTM, but I'm not sure if I understand the high-level need for this.

Specifically, is there any reason for these test streams to hit the broadcaster timeouts but that not happening for production streams? Does the test streams have anything different than what we see in prod?

The link on livepeer/stream-tester#144 to the discord message seems to be broken (or at least I didn't manage to open it) so I couldn't find the original discussion about this.

[victorges]

Should this value be any different than the default?

[leszko]

It's the same, but anyway I think it's good to set it up here, to make sure that it's still ok when the default changes.

[leszko]

LGTM, but I'm not sure if I understand the high-level need for this.

Specifically, is there any reason for these test streams to hit the broadcaster timeouts but that not happening for production streams? Does the test streams have anything different than what we see in prod?

Test Streams are different than the production. In a sense, that even if you're not "good enough" to transcode the data in production, you still want to see your test stream data.

For example, we have Os which have seg transcoding latency > 2s. Then, all Bs requests times out. And now:

• For production => this is a good behavior, we don't want to use this O, because it's too slow; then B needs to fail over to another O
• For test streams => we still do want to see the data for O, just the data will show that the latency is not that good; currently we don't see ANY data for such O, which makes it hard to reason (both for us and for O operator) if there is some problem with test streams or connectivity or it's just that O is slow

[victorges]

Makes a lot of sense! Thanks for explaining