Summary

I noticed that the Kotlin SDK now rejects a non-positive ready polling interval (f494483) because a no-op sleep turns the health check into a spin, and went looking at the Python SDK. ReadinessBudget hands health_check_polling_interval straight to the sleep call, with no check.

With a negative interval the two Python flavours fail differently:

• async: asyncio.sleep() returns immediately for a negative delay, so create / connect / resume / check_ready poll /ping in a tight loop until the deadline. Against a mock transport answering 503, a one-second Sandbox.connect sent 5 pings with the default 200 ms interval and about 3,500 with timedelta(milliseconds=-1).
• sync: time.sleep() raises ValueError, which comes out as SandboxInternalException: Failed to resume sandbox: sleep length must be non-negative. For SandboxSync.create that happens after the sandbox has been provisioned, so it gets killed again because of a client-side argument.

This change raises InvalidArgumentException for a negative interval in ReadinessBudget, and checks it before any request in create (only when the health check is not skipped, since the interval is unused otherwise) and in resume (which calls the resume API before building the budget). connect and check_ready are covered by the budget itself, before any request.

After review I applied the same check to the JavaScript SDK, which passed healthCheckPollingInterval straight to setTimeout. Node clamps a negative delay to about 1 ms, so a one-second Sandbox.connect against a health check returning false made 67 attempts with -1. create (when the health check is not skipped), connect, resume and waitUntilReady now throw InvalidArgumentException before any request. The JS pool options already rejected negative values.

I deliberately left a zero interval allowed: the existing tests call check_ready(..., polling_interval=timedelta(seconds=0)), so I only reject values that never worked in sync mode.

Testing

• Not run (explain why)
• Unit tests
• Integration tests
• e2e / manual verification

New tests/test_readiness_polling_interval.py (16 cases, async and sync, create/connect/resume, asserting no HTTP request is sent). All 14 rejection cases fail on main and pass with the change; the full Python SDK suite passes (625 tests), and ruff check and pyright on the touched files are clean.

New tests/sandbox.polling-interval.test.mjs (create/connect/resume rejections plus a skipped health check case). The three rejection cases fail on main (health check timeout after 67 attempts) and pass with the change. I couldn't run pnpm install on my machine, so I ran the JS tests from the TypeScript sources with Node's type stripping and a stub for openapi-fetch: the new file plus the create, connect/resume, endpoint-readiness and pool suites give 71 passing tests. ESLint was not run locally.

Breaking Changes

• None
• Yes (describe impact and migration path)

A negative interval already crashed the sync SDK; in async it only produced a busy loop. Zero and positive intervals behave as before.

Checklist

• Linked Issue or clearly described motivation
• Added/updated docs (if needed)
• Added/updated tests (if needed)
• Security impact considered
• Backward compatibility considered

AI tools used