[cloudsmith-iduffy]

Description

This PR stacks on #538. #538 added a default timeout to API requests, but a POST kept only the connect timeout. The reason: the session retries a read timeout for every method, so a read timeout on a POST that the API already processed would send it again. The result is a duplicate entitlement, token or package.

So a POST to an API that takes the request body and never replies still waits forever. For example, cloudsmith entitlements create acme/repo --name ci against a server that reads the POST and stays silent:

Before:

STILL HANGING after 90s
server saw: 1 x POST /entitlements/acme/repo/

After:

Creating ci entitlement for the repo repository ... ERROR
Failed to create the entitlement! (status: 0 - Unknown Status)

Detail: ReadTimeout
HTTPConnectionPool(host='127.0.0.1', port=28766): Read timed out. (read timeout=120)

The command fails after 120 seconds, and the server sees the POST one time.

This PR makes two changes:

• POST and PATCH requests get a 120-second read timeout. The API CDN returns a 504 after 60 seconds without a reply, so a slow but healthy request (package create, copy, move, resync) gets a response well before the CLI stops it.
• The retry policy does not retry a POST or PATCH after a read error. urllib3 counts a read timeout and a dropped connection after the request was sent as read errors. In both cases the API can have processed the request. Connect errors still get retries for every method, because the request never left the client. GET, PUT and DELETE keep their read retries.

The dropped-connection case had the same duplicate problem. When the server reads the POST and closes the connection, the CLI sent the POST 7 times in 62 seconds before this change. Now it sends it once and fails at once.

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation update
• Refactoring
• Other (please describe)

Additional Notes

• The read-error rule is in the shared retry policy, so it also applies to other POSTs that use create_requests_session: the OIDC exchange, the SAML refresh and 2FA exchange, and the S3 upload POST. None of these lose a working retry. A retried refresh or TOTP exchange uses a token that the first request already spent. A retried S3 upload POST sent the full Content-Length with an empty body, because the multipart stream was already read. I checked the upload case against a local server.
• A 5xx response to a POST still gets retries, as before. Against a server that reads the POST and returns 504, the CLI sends the POST 7 times. A 504 from the CDN while the API still processes the POST can therefore create duplicates. This PR does not change that. It needs a separate decision about status retries for POST.
• Not breaking. A POST that hung forever now fails with an error. A PATCH that timed out at 30 seconds and got retries now waits 120 seconds one time.
• The change in RetryWithCallback.increment is two lines after the certificate check. It merges cleanly with the increment change in fix(no-ticket): stop retrying unknown hosts and show backoff retries #542.

Test: pytest cloudsmith_cli -m "not integration". The new test runs a local server that reads each request and never replies. It counts one POST and three GETs with retries=2.

🤖 Generated with Claude Code

[Copilot]

🟢 Approval recommended

Static review found no blocking issues in the focused timeout and retry changes; tests were not executed during this review.

No open issues 🎉

What changed in this PR

Builds on #538 to bound stalled API requests without replaying POST or PATCH requests after a read error.

Changes:

• Sets a 120-second default read timeout for POST and PATCH.
• Stops read-error retries for those methods while preserving connection-error and status-code retries.
• Adds local-server regression tests and updates release notes.

File	Description
cloudsmith_cli/​core/​tests/​test_rest.py	Tests timeout defaults and POST-versus-GET read retries.
cloudsmith_cli/​core/​session.py	Prevents POST and PATCH retries after read errors.
cloudsmith_cli/​core/​rest.py	Applies the longer POST and PATCH read timeout.
CHANGELOG.md	Documents bounded waits and revised retry behavior.

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

The changelog inaccurately describes upload behavior, and regression tests leave PATCH, dropped connections, and preserved retry behavior unverified.

2 open issues

• Clarify that upload timeout behavior is unchanged, not upload retries · New
• Test POST and PATCH retries across read timeouts and dropped connections · New

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

No active review issues block approval of the POST/PATCH timeout and retry changes.

No open issues 🎉

2 resolved since last review

• Test POST and PATCH retries across read timeouts and dropped connections
• Clarify that upload timeout behavior is unchanged, not upload retries

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

There are no active review issues for the POST/PATCH timeout and retry changes.

No open issues 🎉

_{Review effort: Balanced}