Uplink timeouts/connectivity issues reported as "no such package available"

[edmorley]

Describe the bug
If the uplink request to a registry fails with a connectivity error such as socket hang up or ETIMEDOUT, the response to the client is an HTTP 404 which makes it give a misleading error message about the package not existing.

The yarn client reports:

error Couldn't find package "align-text@^0.1.3" required by "center-align@^0.1.1" on the "npm" registry.
...
Error: Couldn't find package "har-schema@^2.0.0" required by "har-validator@~5.0.3" on the "npm" registry.
    at MessageError.ExtendableBuiltin (/home/travis/.yarn/lib/cli.js:243:66)
    at new MessageError (/home/travis/.yarn/lib/cli.js:272:123)
    at PackageRequest.<anonymous> (/home/travis/.yarn/lib/cli.js:36403:17)
    at Generator.throw (<anonymous>)
    at step (/home/travis/.yarn/lib/cli.js:98:30)
    at /home/travis/.yarn/lib/cli.js:111:13
    at <anonymous>
    at process._tickCallback (internal/process/next_tick.js:182:7)
...

However the Verdaccio logs show the real cause:

 info <-- 127.0.0.1 requested 'GET /har-schema'
 info --> making request: 'GET https://registry.yarnpkg.com/har-schema'
...
 info <-- 127.0.0.1 requested 'GET /align-text'
 info --> making request: 'GET https://registry.yarnpkg.com/align-text'
...
 http --> ERR, req: 'GET https://registry.yarnpkg.com/har-schema', error: socket hang up
 http <-- 404, user: undefined(127.0.0.1), req: 'GET /har-schema', error: no such package available
...
 http --> ERR, req: 'GET https://registry.yarnpkg.com/align-text', error: ETIMEDOUT
 http <-- 404, user: undefined(127.0.0.1), req: 'GET /align-text', error: no such package available

To Reproduce
Steps to reproduce the behavior:

1. Open a PR against master of https://travis-ci.org/mozilla-neutrino/neutrino-dev thereby triggering the TravisCI run
2. Retrigger the TravisCI job until it fails due to their current connectivity flakiness
3. Look at the Verdaccio logs

Expected behavior

• (ideally, but not the primary issue here) the initial request to the registry should be retried
• if the retry fails, Verdaccio should reply with an HTTP 504 ("Gateway timeout") or some other HTTP status code that would let the yarn client show the correct error message

Configuration File (cat ~/.config/verdaccio/config.yaml)

store:
  memory:
    limit: 10000
uplinks:
  yarnpkg:
    url: https://registry.yarnpkg.com/
    timeout: 20s
    max_fails: 5
packages:
  'neutrino':
    access: $anonymous
    publish: $anonymous
  '@neutrinojs/*':
    access: $anonymous
    publish: $anonymous
  '**':
    access: $anonymous
    proxy: yarnpkg
logs:
  - {type: stdout, format: pretty, level: debug}

Debugging output
https://travis-ci.org/mozilla-neutrino/neutrino-dev/jobs/385364189#L989

Additional context
This issue also occurred when using the npmjs.org registry instead of Yarn's.

[juanpicado]

I will check it our next week, now I am focused in v3 release 🙃. Unless someone around here wants to dive into it.

[jharris4]

I'm having all sorts of issues like this all of a sudden:

{"name":"verdaccio","hostname":"verdaccio1","pid":28733,
"sub":"in","level":35,"request":{"method":"GET","url":"/lerna"},
"user":"jharris","remoteIP":"192.168.124.5 via 192.168.11.31",
"status":404,"error":"File not found",
"bytes":{"in":0,"out":32},"msg":"@{status}, user: @{user}
(@{remoteIP}), req: '@{request.method} @{request.url}', error: @{!error}",
"time":"2018-05-30T22:12:21.389Z","v":0}

It's changing every time I run yarn install, and I think it might be the same issue as this one...

[jharris4]

So, strangely enough I'm not getting these errors any more after I re-installed our verdaccio server but with root ownership of /usr/lib/node_modules (linux).

[edmorley]

I think NPM have made further changes post Cloudflare migration, since the errors I was seeing on Travis (even on GCE) have now stopped with no changes on my part.

[edmorley]

Hi! We're seeing this quite a lot again the last week or two on TravisCI.

Whilst the underlying issue (network connectivity problems/...) aren't Verdaccio's fault - it would be good for Verdaccio to report the real error and/or handle this case better.

[lock]

🤖This thread has been automatically locked 🔒 since there has not been any recent activity after it was closed.
We lock tickets after 90 days with the idea to encourage you to open a ticket with new fresh data and to provide you better feedback 🤝and better visibility 👀.
If you consider, you can attach this ticket 📨 to the new one as a reference for better context.
Thanks for being a part of the Verdaccio community! 💘