Retry downloads if connection terminated mid-response

[dmichon-msft]

pnpm version: 5.9.0

Code to reproduce the issue:

Happens under load on an Azure DevOps Pipeline Agent. Force terminating a connection mid-download in any fashion should repro.

Sometimes connections will get dropped during a download, but after the success headers have been sent. Would be helpful if the fetch wrapper could assume that a JSON parse failure in a registry info fetch or a tarball size mismatch was due to such an issue and retry, rather than terminating the entire process.

Expected behavior:

 WARN  GET https://<REDACTED>/npm/registry/image-size error (Unexpected end of JSON input). Will retry in 10 seconds. 2 retries left.

Actual behavior:

 ERROR  invalid json response body at https://<REDACTED>/npm/registry/image-size reason: Unexpected end of JSON input
at                ../../../../.rush/node-v10.22.1/pnpm-5.9.0/node_modules/pnpm/lib/node_modules/node-fetch-unix/lib/index.js:253  return Body.Promise.reje…
at _tickCallback  internal/process/next_tick.js:68                                                                                                         

The command failed:
 /Users/runner/work/1/common/temp/pnpm-local/node_modules/.bin/pnpm install --store /Users/runner/work/1/common/temp/pnpm-store --no-prefer-frozen-lockfile --recursive --link-workspace-packages false
ERROR: Error: The command failed with exit code 1

Additional information:

• node -v prints: 10.22.1
• Windows, OS X, or Linux?: OS X

[zkochan]

a tarball size mismatch was due to such an issue

this one should already be retried because we check the size of the tarball and also the actual checksum

regarding the JSON, I'll have to look into it. I thought I already worked on this.

Do you commit pnpm-lock.yaml to the repository? If you would, pnpm would not fetch the metadata jsons at all

[zkochan]

🚢 5.11.1