Repository navigation
T146 — The hostname proxy comes back by itself, end to end - #3039
norrietaylor merged 6 commits into
Conversation
The activate names: line now falls back to the ports the VM host daemon serves on when the guest reported none, as min ls does.
The case holds the proxy port, activates a box, checks the cause once at activate and on min ls, frees the port, and opens the box by name through the proxy without a restart. On a VM lane it pins MINVMD_NODE_PROXY_PORT.
…me-proxy-comes-back-by-itself-end-to-e
|
Important Review skippedBot user detected. To trigger a single review, invoke the
|
|
Approved: no findings at Nothing to do. Verdict APPROVE, posted as a comment: the App cannot review its own pull request; the check and the label carry it. Clobber: none seen. |
There was a problem hiding this comment.
arch-review: PASS · the full ruling
The change conforms to §7.1 of the networking design. The names: line now takes the port from the VM host daemon, not the guest. It uses the same read as min ls, so line 244 holds. A down proxy still names its cause and no port, as line 254 requires. The new e2e case pins the port through the configured surfaces of line 264. The recovery half did not run on the author's host. The required KVM lane runs it with the switch set.
Ruled at 8ca829a against gominimal/arch@7a86ec7bb3c0 by claude-opus-5-5.
|
Round 1 of 10 on a failed check suite: the coder changed nothing. Round 1 changes no code. I diagnosed the failure, and the fix is outside the task's layers.
Cause: on a VM host whose live name surface is the proxy, crates/minimal/src/cmd/list.rs folds the VM host daemon's proxy-down cause into the BOX NAMES row (resolver::surface_beside_proxy_down gives LiveSurface::ProxyNotServing). resolver::proxy_down_line_beside then returns None, so the HOSTNAME PROXY row does not print. No 'listening on' row prints either, because serving_proxy_port is None and the VM serves no ports. NET-020 r…[truncated] |
|
/revise Script only: change
|
…me-proxy-comes-back-by-itself-end-to-e
On a VM host min ls prints the proxy cause on the BOX NAMES row and no HOSTNAME PROXY row while the port is held (#3046). The native lane keeps the HOSTNAME PROXY check. After the recovery the case checks that no row names the cause and that HOSTNAME PROXY names the port.
There was a problem hiding this comment.
arch-review: PASS · the full ruling
The new commit changes only the e2e case. On a VM lane the case reads the proxy cause from the BOX NAMES row, and it now requires that min ls prints the cause once. Section 7.1 line 254 requires only that a failed bind is a surfaced error, and the case still proves that on both lanes. The row difference from NET-020 is in code that #3005 owns, and #3046 tracks it. Add #3046 to the PR body as a residual. The VM leg did not run yet. CI on the KVM and macOS lanes is its proof.
Ruled at af0151a against gominimal/arch@7a86ec7bb3c0 by claude-opus-5-5.
On a native developer host with no box sandbox, a host-address box gets no PID namespace of its own, so every exec into it fails. The case now gates the responder step on an exec, as host_address_names_and_host_alias does: NOT RUN on a native developer run, a failure on CI or a VM lane.
There was a problem hiding this comment.
arch-review: PASS · the full ruling
This push adds one gate to the e2e case. The gate records NOT RUN when a native developer host cannot run a box, and it fails the case on CI and VM lanes. That copies the gate in proof_host_address_names_and_host_alias, and the CI legs stay hard. The design does not govern e2e skip gates, and the compare does not touch any other part of the change. The PASS on af0151a7 stands for this head.
Ruled at 2e1c9b2 against gominimal/arch@7a86ec7bb3c0 by claude-opus-5-5.
|
Round 2 of 10 on a /revise note, pushed as
Verify On a VM lane the case now reads the held-port cause from the BOX NAMES row of min ls. It checks the port, the holder and that the cause prints once. A native lane keeps the HOSTNAME PROXY check. After the port frees, the case checks that no min ls row names the cause and that the HOSTNAME PROXY row names 127.0.0.1: as listening. Notes from the agent: Script-only round, as the /revise note asked. Changed only scripts/session-e2e.sh. (1) On minvmd the min ls check reads 'BOX NAMES:', with the same checks for 'the hostname proxy is not serving', 127.0.0.1:$prf_pin and the holder. It also checks that the cause appears once in the min ls output. (2) The native lane keeps the HOSTNAME PROXY check as it was. (3) The case comment says the VM row differs from NET-020 and links #3046. The ::error:: texts name the row the lane reads. (4) I read the later checks against the VM path. In list.rs, serving_proxy_port is None while proxy_down is set. surface_beside_proxy_down moves the cause onto BOX NAMES, so no HOSTNAME PROXY row prints while the port is held. The 'listening on' check during the hold stays valid. After the recovery, proxy_down clears and reported_proxy_port falls back to the VM host daemon's served ports, so HOSTNAME PROXY names the port. The post-recovery check now fails if any row still names the cause, not only the first HOSTNAME PROXY row. It retries min ls for up to 10 s, so the listing can catch up with the proxy. The second activate reads the same fallback in create.rs. The minvmd record 'the hostname proxy is serving after its port freed' is at info level in proxy_reclaim.rs, and the case's RUST_LOG keeps minvmd=info. The recovery half is still not run on any lane: this box has no /dev/kvm, and its native run stopped at the box spawn because of the nested-sandbox limit. |
There was a problem hiding this comment.
Approved at 2e1c9b2. proxy_recovers_when_its_port_frees ran to the end on both VM lanes: the box started with the port held, the cause printed once at activate and on min ls, and the box answered by name about a second after the port freed, with no restart. The names: line now falls back to the port the VM host daemon serves on, as min ls does. The BOX NAMES row on VM lanes differs from NET-020, and #3046 tracks it. CI, arch-review and minimal-review pass.
T146 — The hostname proxy comes back by itself, end to end
slice: S23
kind: integration
covers: none
size: M
est: 60 code, 200 tests
executor: agent
difficulty: routine
model: standard
diagnostics: a person reads the case's output, which prints the activate and
min lslines it checked while the port was held and the proxied request's status after it freed, and the VM host daemon's retry lines in the case's work directoryobservability: the case prints each step with the holder, the line it matched and the time the proxy took to serve after the port freed
layers:
- e2e:
- scripts/session-e2e.sh: adds the case proxy_recovers_when_its_port_frees: holds the hostname proxy's port with a listener, activates a box, checks the cause prints once at activate and on
min ls's HOSTNAME PROXY row, frees the port, and opens the box by name through the proxy without a restart; withE2E_VMset it pins the proxy port so the VM starts with its proxy down- crates/minimal/src/cmd/boxes/create.rs: the activate
names:line reads the port the VM host daemon recovered, asmin lsdoestest: ./scripts/session-e2e.sh proxy_recovers_when_its_port_frees
test: cargo nextest run -p minimal listener_failure_reported_with_remedy
produces: none
consumes: none
depends on:
- blocked by #2942
- blocked by #2950
proof: When another program holds the hostname proxy's port, boxes and their VM still start,
min session activatenames the holder once andmin lsnames it on its HOSTNAME PROXY row, and box names open through the proxy without a restart once the port is free.Proof
With another program on the hostname proxy's port, a box still starts. The new e2e case proxy_recovers_when_its_port_frees checks that
min session activatenames the cause once on its names: line and thatmin lsnames it on its HOSTNAME PROXY row. On a VM lane the check includes who holds the port. The case then frees the port and checks that the box opens by name through the proxy on the same port, with no restart. A box activated after the recovery names that port on its names: line, because activate now falls back to the ports the VM host daemon serves on, asmin lsdoes. On this host only the held-port half was observed. The recovery half did not run, because this host cannot start box processes.Notes from the agent: Observed on this host (native lane, no /dev/kvm): the activate names: line and the min ls HOSTNAME PROXY row both named the held port 18756 once. The case stopped at the first
min session exec: 'Minimal cannot read the box's process namespace'../scripts/session-e2e.sh session_execfails the same way here, so the cause is the environment and not this case. The VM branch (E2E_VM, MINVMD_NODE_PROXY_PORT pin, minvmd retry records copied to $WORK/prf-retry.log) did not run anywhere. It needs the KVM or macOS lane with MINVMD_GVPROXY_BIN. Without a switch the VM branch reports NOT RUN. create.rs: the names: line now uses crate::cmd::list::reported_proxy_port with the ports from crate::cmd::session::vm_host_proxy_ports_at. That read runs only on a VM-backed host when the guest reported no port. The case is also in the whole-lane order, after hostnames_recover_and_two_daemons_route, and it stops both daemons at its end so that no later case inherits the pin. The added PRF_HOLD_PID/PRF_SEED_DIR globals and teardown lines are in scripts/session-e2e.sh.Tests
./scripts/session-e2e.sh proxy_recovers_when_its_port_freescargo nextest run -p minimal listener_failure_reported_with_remedyVerify
min task run agent-verifyexited 100 under rustc 1.99.0, which the repository does not pin (1.99.1 in rust-toolchain.toml); the PR is not a draft on that alone — CI, which honours the pin, is the gate.Base
The base branch moved while the coder worked and was merged in before the tests ran. These files changed on BOTH sides since the branch point — read them first, because a rewrite from a stale reading can undo a merge without a conflict:
scripts/session-e2e.shCloses #2952