runtime-split.md

Runtime split: pod-run evidence

This file records the runtime-split pod checks on two clusters, and the openab-pty spec hash before and after the runs. Source: REPORT.md and the pod-run evidence logs in the runtime-split task folder.

OrbStack arm64

Check	Result	Exit code or observed result
3(a)	PASS	Pod Running after the manifest fix (see First runs). Init containers volume-prep, netinit, expiry-store-bootstrap exit 0. agent, proxy, runtime, egress ready, restarts 0.
3(b)	PASS	Proxy listener HTTP 403. Egress listener HTTP 405. Other loopback port refused for UID 1000, curl exit 7. Direct TCP to upstream curl exit 7. DNS lookup rc 1.
3(c)	PASS	Register 201, surrogate prefix aps_. Granted request 200. Unknown surrogate 403. Revoke 204, then 403. Runtime hello 201 in proxy audit. Upstream POSTs 2 before recreate, real credential seen, surrogate and caller key not seen.
3(d)	PASS	Secret hits 0 in 39 log lines. 0 in 41 lines after recreate. 0 in 40 lines of overlay logs.
3(e)	PASS	Store sha256 prefix 779620227f8b1122, 333 bytes, 5 lines. Identical before delete, after delete, after recreate.
3(f)	PASS	Env hits 0, process hits 0, filesystem scan hits 0.
A(i)	not recorded	not recorded
A(i-a)	PASS	Connect to session.sock as UID 1000: ENOENT.
A(i-b)	PASS	Connect to the same socket path through the test-only overlay: EACCES. Socket owner:group and mode 1001:1002 660.
A(ii)	PASS	POST /v1/hello and GET /v1/hello/challenge both 404 unknown_operation.
A(iii)	PASS	Rerun (see Reruns). 60 samples over about 4 min. session.sock EACCES in all 60. Runtime hello never 201. Proxy audit: 201, 409 authority_active three times, 201. Runtime restarted 4 times.

The first pass of 3(e) on this cluster printed the store fields wrong (a parsing bug, per REPORT.md). Its byte and line values are not used. The table shows the valid rerun only.

Second node amd64

Check	Result	Exit code or observed result
3(a)	PASS	Pod Running, init containers exit 0, app containers ready, restarts 0. Netinit rules for UID 1002 present in the log.
3(b)	PASS	Proxy listener HTTP 403. Egress listener HTTP 405. Other loopback port refused, curl exit 7. Direct TCP to upstream curl exit 7. DNS lookup rc 1.
3(c)	PASS	Register 201, granted request 200, unknown surrogate 403, revoke 204 then 403. Runtime hello 201 in audit. Upstream POSTs 2 before recreate, real credential seen, no surrogate or caller key seen.
3(d)	PASS	Secret hits 0 in 39 log lines. 0 in 41 lines after recreate. 0 in 40 lines of overlay logs.
3(e)	PASS	Rerun after two FAIL attempts (see First runs). Store sha256 prefix 3fb29a547c722a07, 181 bytes, 3 lines. Identical before delete, after delete, after recreate.
3(f)	PASS	Env hits 0, process hits 0, filesystem scan hits 0.
A(i)	not recorded	not recorded
A(i-a)	PASS	Connect to session.sock as UID 1000: ENOENT.
A(i-b)	PASS	Connect to the same socket path through the overlay: EACCES. Socket owner:group and mode 1001:1002 660.
A(ii)	PASS	POST /v1/hello and GET /v1/hello/challenge both 404 unknown_operation.
A(iii)	PASS	60 samples over about 4 min. session.sock EACCES in all 60. Runtime hello never 201. Proxy audit: 201, 409 authority_active three times, 201. Runtime restarted 4 times after the kill.

First runs (FAIL)

Cluster	Check	Result	Exit code or observed result
OrbStack, first run	3(a)	FAIL	Pod stayed Pending in Init, and volume-prep restarted. chmod returned EPERM after chown, because the manifest gave CHOWN as the only capability. The netinit rules check also FAILed in this run; the source gives no separate cause. Fixed by adding FSETID and running chmod before chown. Rerun: PASS (see Reruns).
OrbStack, first run	A(iii)	FAIL	60 samples; session.sock EACCES in all; runtime hello never 201; restarted=False. Reason: the probe sent kill -9 to PID 1 of the runtime container. The kill returned rc 0 and did not stop PID 1, so no restart could be seen. This run also recorded proxy_hello_only_runtime=False; the logs do not say why. Rerun: PASS (see Reruns).
Second node, first run	3(e)	FAIL	The log shows two FAIL attempts (REPORT.md mentions one). Both printed the store sha256, bytes and lines as ?, so no value was read. Reason as recorded: the reader pod was not ready, and the store-read command had a quoting bug. The harness was then changed to wait for the reader pod and to quote the command. Rerun: PASS (see Reruns).

Reruns (PASS)

Cluster	Check	Result	Exit code or observed result
OrbStack, rerun	3(a)	PASS	Pod Running after the manifest fix. Init containers exit 0, app containers ready, restarts 0. Netinit rules check PASS.
OrbStack, rerun	A(iii)	PASS	The runtime shim ran as a child of a wrapper shell, and the probe killed that child (killed-pid 6, rc 0). 60 samples; session.sock EACCES in all; runtime hello never 201; restarted=True; restarts 4; proxy audit 201, 409 three times, 201; proxy_hello_only_runtime=True.
Second node, rerun	3(e)	PASS	Store sha256 prefix 3fb29a547c722a07, 181 bytes, 3 lines. Identical before delete, after delete, after recreate.

openab-pty before and after

Cluster	Check	Result	Exit code or observed result
OrbStack	before	PASS	Spec hash 7a3bf0a4...216ad. 1 pod, Running.
OrbStack	after	PASS	Spec hash 7a3bf0a4...216ad, identical to before. 1 pod, Running.
Second node	before	PASS	Spec hash 4e171458...95430. 6 pods. One Pending pod was already Pending before the run and was not touched.
Second node	after	PASS	Spec hash 4e171458...95430, identical to before. 6 pods.

Matrix

Result: 41 of 41, PASS. Run on the build host.

• Exit code: not recorded.
• Observed: 41 rows run, 41 expected. Summary 41 PASS of 41 rows. Gate 28 of 28. Phase 1 13 of 13.