fix(polling): treat --timeout as a wall-clock budget, not an attempt count

[danishashko]

Problem

poll_until looped exactly timeout_seconds times, sleeping the interval (1s) after each poll but never accounting for the request round-trip. So the stated timeout undercounted real time: --timeout 600 ran for roughly 12 minutes instead of 10, because each of the 600 polls added its own fetch time on top of the 1s sleep. This is the "create timed out / failed, but the collector was fine minutes later" surprise.

Fix

Poll until a real wall-clock deadline (Date.now() + timeout_seconds * 1000) instead of counting iterations, and stop before a sleep that would run past the budget. Now the stated --timeout matches real elapsed time regardless of how slow each request is.

Tests

Updated the timeout test to fake timers (vi.useFakeTimers + advanceTimersByTimeAsync) so it verifies the wall-clock behaviour without waiting in real time. tsc clean, full suite passes with no new failures.

[danishashko]

Problem

poll_until looped exactly timeout_seconds times, sleeping the interval (1s) after each poll but never accounting for the request round-trip. So the stated timeout undercounted real time: --timeout 600 ran for roughly 12 minutes instead of 10, because each of the 600 polls added its own fetch time on top of the 1s sleep. This is the "create timed out / failed, but the collector was fine minutes later" surprise.

Fix

Poll until a real wall-clock deadline (Date.now() + timeout_seconds * 1000) instead of counting iterations, and stop before a sleep that would run past the budget. Now the stated --timeout matches real elapsed time regardless of how slow each request is.

Tests

Updated the timeout test to fake timers (vi.useFakeTimers + advanceTimersByTimeAsync) so it verifies the wall-clock behaviour without waiting in real time. tsc clean, full suite passes with no new failures.

[danishashko]

Problem

poll_until looped exactly timeout_seconds times, sleeping the interval (1s) after each poll but never accounting for the request round-trip. So the stated timeout undercounted real time: --timeout 600 ran for roughly 12 minutes instead of 10, because each of the 600 polls added its own fetch time on top of the 1s sleep. This is the "create timed out / failed, but the collector was fine minutes later" surprise.

Fix

Poll until a real wall-clock deadline (Date.now() + timeout_seconds * 1000) instead of counting iterations, and stop before a sleep that would run past the budget. Now the stated --timeout matches real elapsed time regardless of how slow each request is.

Tests

Updated the timeout test to fake timers (vi.useFakeTimers + advanceTimersByTimeAsync) so it verifies the wall-clock behaviour without waiting in real time. tsc clean, full suite passes with no new failures.

[danishashko]

Problem

poll_until looped exactly timeout_seconds times, sleeping the interval (1s) after each poll but never accounting for the request round-trip. So the stated timeout undercounted real time: --timeout 600 ran for roughly 12 minutes instead of 10, because each of the 600 polls added its own fetch time on top of the 1s sleep. This is the "create timed out / failed, but the collector was fine minutes later" surprise.

Fix

Poll until a real wall-clock deadline (Date.now() + timeout_seconds * 1000) instead of counting iterations, and stop before a sleep that would run past the budget. Now the stated --timeout matches real elapsed time regardless of how slow each request is.

Tests

Updated the timeout test to fake timers (vi.useFakeTimers + advanceTimersByTimeAsync) so it verifies the wall-clock behaviour without waiting in real time. tsc clean, full suite passes with no new failures.

[danishashko]

Problem

poll_until looped exactly timeout_seconds times, sleeping the interval (1s) after each poll but never accounting for the request round-trip. So the stated timeout undercounted real time: --timeout 600 ran for roughly 12 minutes instead of 10, because each of the 600 polls added its own fetch time on top of the 1s sleep. This is the "create timed out / failed, but the collector was fine minutes later" surprise.

Fix

Poll until a real wall-clock deadline (Date.now() + timeout_seconds * 1000) instead of counting iterations, and stop before a sleep that would run past the budget. Now the stated --timeout matches real elapsed time regardless of how slow each request is.

Tests

Updated the timeout test to fake timers (vi.useFakeTimers + advanceTimersByTimeAsync) so it verifies the wall-clock behaviour without waiting in real time. tsc clean, full suite passes with no new failures.

[danishashko]

Problem

poll_until looped exactly timeout_seconds times, sleeping the interval (1s) after each poll but never accounting for the request round-trip. So the stated timeout undercounted real time: --timeout 600 ran for roughly 12 minutes instead of 10, because each of the 600 polls added its own fetch time on top of the 1s sleep. This is the "create timed out / failed, but the collector was fine minutes later" surprise.

Fix

Poll until a real wall-clock deadline (Date.now() + timeout_seconds * 1000) instead of counting iterations, and stop before a sleep that would run past the budget. Now the stated --timeout matches real elapsed time regardless of how slow each request is.

Tests

Updated the timeout test to fake timers (vi.useFakeTimers + advanceTimersByTimeAsync) so it verifies the wall-clock behaviour without waiting in real time. tsc clean, full suite passes with no new failures.

[danishashko]

Problem

poll_until looped exactly timeout_seconds times, sleeping the interval (1s) after each poll but never accounting for the request round-trip. So the stated timeout undercounted real time: --timeout 600 ran for roughly 12 minutes instead of 10, because each of the 600 polls added its own fetch time on top of the 1s sleep. This is the "create timed out / failed, but the collector was fine minutes later" surprise.

Fix

Poll until a real wall-clock deadline (Date.now() + timeout_seconds * 1000) instead of counting iterations, and stop before a sleep that would run past the budget. Now the stated --timeout matches real elapsed time regardless of how slow each request is.

Tests

Updated the timeout test to fake timers (vi.useFakeTimers + advanceTimersByTimeAsync) so it verifies the wall-clock behaviour without waiting in real time. tsc clean, full suite passes with no new failures.

[danishashko]

Problem

poll_until looped exactly timeout_seconds times, sleeping the interval (1s) after each poll but never accounting for the request round-trip. So the stated timeout undercounted real time: --timeout 600 ran for roughly 12 minutes instead of 10, because each of the 600 polls added its own fetch time on top of the 1s sleep. This is the "create timed out / failed, but the collector was fine minutes later" surprise.

Fix

Poll until a real wall-clock deadline (Date.now() + timeout_seconds * 1000) instead of counting iterations, and stop before a sleep that would run past the budget. Now the stated --timeout matches real elapsed time regardless of how slow each request is.

Tests

Updated the timeout test to fake timers (vi.useFakeTimers + advanceTimersByTimeAsync) so it verifies the wall-clock behaviour without waiting in real time. tsc clean, full suite passes with no new failures.