fix(exporter-collector): all http export requests should share same a…

[blumamir]

Which problem is this PR solving?

• fixes new http.Agent() is created for each socket in collector exporter instead of using single agent for all requests #1840
  Currently, each new HTTP request in exporter collector is newing a new agent, which means different requests do not share connection pool, and that each one is creating a brand new socket with tcp handshake and TLS handshake.

Short description of the changes

Create a single http/https agent in exporter constructor and set the same one in the options for each HTTP request.
As I wrote in the issue, In my opinion it's better to pass an initialized agent to the collector instead of the collector creating one for itself. Since there was no discussion about it, I implemented the fix to stay as close as possible to the current implementation:

• default is to create an agent for the exporter.
• to disable agent, user need to pass { keepAlive: false } to the exporter options.
• Additional agent parameters can be passed in the httpAgentOptions option.
• In case httpAgentOptions is used, if user set keepAlive: false, a warning will be printed to log and agent will not be used (node's agent allow this configuration, why do we prevent it in the exporter?)

I added a unittest for the change, and tested it locally in a real application, and observed the latency of a single export after warmup. My test application is located in Israel and export spans to collector in aws Ireland. Export time dropped from ~500ms prior to the fix, to ~100ms with the fix.

[codecov]

Codecov Report

Merging #1863 (f57a562) into master (a78bb80) will decrease coverage by 0.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #1863      +/-   ##
==========================================
- Coverage   92.68%   92.66%   -0.02%     
==========================================
  Files         174      174              
  Lines        6040     6040              
  Branches     1284     1284              
==========================================
- Hits         5598     5597       -1     
- Misses        442      443       +1     

Impacted Files	Coverage Δ
...emetry-core/src/platform/node/RandomIdGenerator.ts	87.50% <0.00%> (-6.25%)	⬇️

[blumamir]

The lint fail is not related to my PR:

lerna ERR! npm run docs-test stderr:
🏊‍♂️ crawling docs/out
[404] https://open-telemetry.github.io/opentelemetry-js/classes/traceapi.html
[404] https://open-telemetry.github.io/opentelemetry-js/classes/propagationapi.html
[404] https://open-telemetry.github.io/opentelemetry-js/classes/contextapi.html
docs/out
  [404] https://open-telemetry.github.io/opentelemetry-js/classes/traceapi.html
  [404] https://open-telemetry.github.io/opentelemetry-js/classes/propagationapi.html
  [404] https://open-telemetry.github.io/opentelemetry-js/classes/contextapi.html
ERROR: Detected 3 broken links. Scanned 55 links in 0.48 seconds.

[obecny]

@naseemkullah it should not be merged yet -> please read https://github.com/open-telemetry/opentelemetry-js/blob/master/CONTRIBUTING.md#general-merge-requirements

[naseemkullah]

Oops, sorry @obecny, another maintainer approval was required.

[obecny]

Oops, sorry @obecny, another maintainer approval was required.

it is also a breaking change so I would expect 24 hours wait so more people can have a chance to look at.

[blumamir]

Oops, sorry @obecny, another maintainer approval was required.

it is also a breaking change so I would expect 24 hours wait so more people can have a chance to look at.

why is it a breaking change?

[obecny]

why is it a breaking change?

You have removed public keepAlive and httpAgentOptions from class

[dyladan]

@naseemkullah please when merging ensure one of the changelog labels from lerna.json is applied

[naseemkullah]

@naseemkullah please when merging ensure one of the changelog labels from lerna.json is applied

Thanks @dyladan, will do!

[dyladan]

Additionally, merge requirements were not met for this PR. Merge by a non-maintainer requires at least 2 maintainer approvals. This PR only has 1. I am not going to revert but please keep this in mind in the future.

[naseemkullah]

@dyladan yes true, this was pointed out by @obecny and acked: #1863 (comment) :)

[dyladan]

@blumamir since this is a breaking change can you please document the break in the migration section of the README in another PR?

[blumamir]

@blumamir since this is a breaking change can you please document the break in the migration section of the README in another PR?

sure, I'll send this PR tomorrow.
do you mean to add something like the Upgrade guidelines currently in the README.md?
Should I add 0.15.0 to 0.16.0?

[obecny]

I guess it will be enough to mention what has been removed.

[blumamir]

@blumamir since this is a breaking change can you please document the break in the migration section of the README in another PR?

#1867