fix(tutorials): retry flaky PyPI fetches in pydantic_ai images - #547

What changed

The two 110_pydantic_ai tutorial image builds failed four times when a PyPI download dropped mid-stream in uv pip install ("Failed to fetch … broken pipe" or a TLS EOF), mostly in the emulated arm64 half of the build. uv 0.6.4 doesn't retry that error. One failure hit a publish run, so the 00_base :latest image wasn't updated.

In 10_async/00_base/110_pydantic_ai/Dockerfile and 10_async/10_temporal/110_pydantic_ai/Dockerfile:

• The uv pip install --system .[dev]… step runs up to 3 times, waiting 10 s and then 20 s between attempts. If all 3 fail, the build still fails.
• UV_CONCURRENT_DOWNLOADS=4 (uv's default is 50), so the slow arm64 build opens fewer connections at once.

Not changed: the uv version. All 35 tutorial Dockerfiles pin uv 0.6.4, each on its own line. Bumping two would split the pin, and moving all 35 to 0.12.x is bigger than this fix. That's a separate PR. The publish workflow's platforms are unchanged too.

How this was verified

• Retry loop logic: ran the same loop with a stand-in command:

  Case	Result
  Succeeds first time	exit 0, no retries
  Fails twice, then succeeds	2 "failed (attempt n of 3)" lines, exit 0
  Fails every time	3 failure lines, exit 1, so the build fails

• Image builds: this PR's CI builds both images for linux/amd64 and linux/arm64 and runs their tests. I couldn't build them locally because my local Docker disk was full.

• Not tested: a real mid-download PyPI failure, which can't be forced on demand.

🤖 Generated with Claude Code

Confidence Score: 5/5

Summary

Both 110_pydantic_ai tutorial image builds limit PyPI downloads and retry package installs. This gives temporary download interruptions another chance to recover, especially in the slow emulated arm64 builds.

• Both pydantic_ai tutorial images limit downloads and retry package installs.

_{Reviews (3) · Last reviewed commit: "fix(tutorials): retry flaky PyPI fetches..." · Reviewed by Greptile}