[slayerjain]

s3.sh and s3-test.sh are meant to be one shared pair: every keploy repo that carries them pins their sha256, and a change goes into all of them together. Today the carriers other than keploy/enterprise and playwrightDB have the previous pair (the "before" column). The final pair is on keploy/enterprise's branch train/fix/s3-op-timeout-by-progress-next (46c66c5e5), which is to become keploy/enterprise#2929's head; keploy/playwrightDB#37 is to move playwrightDB to it. This PR copies the new shared s3-test.sh into keploy-ci-node byte for byte and moves the pin in .github/workflows/docker-build.yml (want_test) to its hash. s3.sh does not change.

file	before	after
s3.sh	40da95b471bb8e77f7851b2a32b04317b3743f3a289b730b10caf01145e8fbac	unchanged
s3-test.sh	bb41f8ad34ed3a4e5fc8c225214b98fedf778cc2b780074dfe44681c0580b268	a10f2713f30cc6b42c8be68c50fcce099aae5718178dbd3b7261637760b38321

What the suite now catches

• The fake clock is checked. Two cases move it themselves: stdin that arrives 4 s late to put, and a reader that stops for 3 s on cat. Nothing checked that the clock moved. With that step a no-op, every case still passed, and so did the two s3.sh mutants those cases are there for: the first request not counted as progress, and the reader not counted as progress. Each case now checks the clock it ends at, 1004 and 1003. Neither case makes a backoff, so the value is exact.

• S3_OP_TIMEOUT's deadline refusal is reached. FAKE_SLEEP_LATE=N makes the stub sleep end N s late, as a sleep does on a loaded box. Two new cases use it:

  • once the deadline is reached, no request is made (exit 5, one call, the clock at 1003);
  • with 1 s left, the request is still made, its connect timeout cut to 1 s.

  No case reached that refusal before. With it removed, or with its check made -ge 0, every case passed.

• The fake clock can't be read half-written or set from outside. The clock file is written whole (a new file renamed over it), so a reader never finds it empty. FAKE_CLOCK and FAKE_SLEEP_LATE are unset before the stubs are written, so the runner's environment can't set them.

Verification

• The commands of the "Every baked s3.sh is the shared copy" step, on this tree: exit 0.
• s3-test.sh hermetic on an amd64 box: ALL PASS under sh and bash.
• Failing first, on the old file (bb41f8ad, with s3.sh 40da95b4), under bash, sh, dash and busybox ash: with the two cases' own clock step a no-op, every case passes, alone and with each of the two s3.sh progress mutants; with the deadline refusal removed, or its check made -ge 0, every case passes too.
• The new file was proven in keploy/enterprise's tree, with the same s3.sh and s3-test.sh bytes as 46c66c5e5 (the commit that is to become keploy/enterprise#2929's head, on branch train/fix/s3-op-timeout-by-progress-next; the runs were on main 591b20c16, before a rebase that changes neither file). A tick that is a no-op, one second short or one second long fails both clock cases under bash, dash and busybox ash, and so does a no-op tick with each s3.sh progress mutant. The two deadline mutants fail under bash, sh and dash.

Order

Every carrier takes this pair before keploy/enterprise#2929 merges, and keploy/enterprise#3067 tracks them. No carrier's guard reads another repo's copy, so this PR is green on its own.