Compare commits

...

4 Commits

Author SHA1 Message Date
Dai Ha cfebc575ea fleetd #450: make PeerLauncher.spawn(SpawnRequest, PlacementDecision) abstract
CI / contract (pull_request) Successful in 47s
CI / build (pull_request) Successful in 2m5s
The default re-entered the single-argument spawn(SpawnRequest), which re-runs
checks that can refuse the profile place() just chose (#444's window). Only
CompositePeerLauncher overrode it; a future placement-doing launcher could
have inherited the wrong body silently.

Give every current implementer an explicit override, chosen by what it does:
- HerdrPeerLauncher (base of ClaudeCodeLauncher/OpenCodeLauncher, neither of
  which overrides spawn(req) or place()) does no placement filtering of its
  own, so it gets the re-entering form.
- CompositePeerLauncher's routing-form override is untouched.
- 5 test-fake PeerLauncher implementers (SessionManagerTest, FleetdBackendErrorSinkTest)
  get overrides matching their existing spawn(SpawnRequest) shape: delegating
  wrappers delegate, unreachable stubs throw, the single-profile fake re-enters.

ConfigRef does not implement PeerLauncher at all (confirmed in this tree at
822327e) despite the ticket listing it as a src/main implementer.
2026-09-10 17:10:44 +07:00
ltms 822327eed5 Merge #447: pin the place()-to-spawn() window PlacementDecision closes (fleetd #444)
CI / contract (push) Successful in 1m33s
CI / build (push) Successful in 1m36s
Verified by the lead on the exact tree that lands (head e4c703a, base 82fae94 is
an ancestor, so this is the tree I measured):

  FULL BUILD  Tests run: 1575, Failures: 0, Errors: 0, Skipped: 0  BUILD SUCCESS
              compile errors: 0
  CONTROL     CompositePeerLauncherTest  Tests run: 76, Failures: 0  -> GREEN

  M1  the 2-arg spawn re-enters the 1-arg spawn (the inherited default this
      ticket forbids for a multi-profile launcher)
      -> KILLED  Errors: 1
      CompositePeerLauncherTest
        .spawnHonorsAPlacementDecisionEvenAfterItsProfileIsQuarantinedInTheWindowAfterPlace

  M2  drop the stamping: route to the decided profile but do not carry it
      (SpawnRequest routedReq = req)
      -> KILLED  Failures: 1
      same test method

Both mutations proved applied by printing the mutated method, and the tree was
restored clean after each (git status --porcelain empty).

Round 1 of this PR had a fixture weakness I found by mutation: StubLauncher's own
fallback default was "sol", the same profile place() decides, so an unstamped
request landed on spawnCount("sol") by coincidence and M2 survived. e4c703a gives
the adapter "b" as its fallback instead. One fixture now kills both mutations.

src/main is javadoc-only in this PR: 12 added lines, 0 added code lines, measured
by filtering the main-side diff.
2026-09-10 12:00:59 +02:00
Dai Ha e4c703a51a fleetd #444: separate the adapter's fallback default from the decided profile
CI / contract (pull_request) Successful in 46s
CI / build (pull_request) Successful in 1m56s
Review found the fixture's StubLauncher fell back to 'sol' too — the
same profile place() decides — so an UNSTAMPED request could land on
spawnCount('sol') by coincidence, and the assertion's claim that the
request 'actually carried sol' was unproven. Dropping the stamping
(SpawnRequest routedReq = req) while keeping the routing survived the
test unchanged.

Fix: give the adapter 'b' as its own fallback default instead, so an
unstamped request counts against 'b', not 'sol'. Verified both
mutations against the single test in isolation:
  - drop-stamping (routedReq = req): RED, expected <sol> but was <b>
  - re-entering (return spawn(req.withProfile(decision.profile())))
    i.e. M1 from the first round: still RED, PlacementException
    naming the now-quarantined 'sol'
Restored both; full suite green at 1575 tests.

No changes to src/main — PeerLauncher's javadoc from the first round
is unchanged.
2026-09-10 16:50:39 +07:00
Dai Ha 3f036b2a62 fleetd #444: pin the place()-to-spawn() window PlacementDecision closes
CI / contract (pull_request) Successful in 1m4s
CI / build (pull_request) Successful in 1m53s
Add a test that resolves place(role) while nothing is quarantined,
then quarantines the resolved profile's credential BEFORE spawning
against the held PlacementDecision. CompositePeerLauncher.spawn(req,
decision) must still honor the decision and land on the quarantined
profile, since it never re-runs the explicit-profile enforce* checks.

Verified the test kills the regression: with the override's body
replaced by the re-entering spawn(req.withProfile(...)) form, this
exact test goes RED with a PlacementException naming the now-
quarantined profile; restored, the full suite is green (1575 tests).

Also documents on PeerLauncher's default spawn(req, decision) that a
launcher routing across more than one profile MUST override it,
naming the four enforce* checks the default's re-entry re-applies.
2026-09-10 16:41:52 +07:00
5 changed files with 126 additions and 6 deletions
@@ -17,6 +17,7 @@ import dev.ltms.fleet.peer.PeerHandle;
import dev.ltms.fleet.peer.PeerLauncher;
import dev.ltms.fleet.peer.PeerUnreachableException;
import dev.ltms.fleet.peer.SpawnRequest;
import dev.ltms.fleet.placement.PlacementDecision;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
@@ -594,6 +595,23 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
req.sessionName(), spawned.agentSessionId(), spawned.receipt());
}
/**
* {@inheritDoc}
*
* <p>fleetd #450: re-enters {@link #spawn(SpawnRequest)} with {@code decision}'s profile named
* explicitly. This is the re-entering form the interface javadoc describes for a launcher with
* no placement concept of its own — an instance of this class spawns a single adapter's own
* profile set by explicit name only ({@link #place}/{@link #defaultProfileFor} are unoverridden
* here and just wrap {@link #defaultProfile()}); it does no quarantine/cool-off/maxLoad/model-off
* filtering of its own to re-apply. That filtering lives one layer up, in {@code
* CompositePeerLauncher}, which is the launcher that routes across more than one profile and
* therefore overrides this method with the routing form instead.
*/
@Override
public PeerHandle spawn(SpawnRequest req, PlacementDecision decision) {
return spawn(req.withProfile(decision.profile()));
}
/** The herdr daemon that owns this launcher's pane coordinates. */
public HerdrClient herdr() {
return agents.herdr();
@@ -253,15 +253,27 @@ public interface PeerLauncher {
* matching profile; passing a request that names a <em>different</em>, explicit profile than
* the decision it is paired with is a caller bug this method does not attempt to detect.
*
* <p>Default implementation for a launcher with no placement concept of its own: delegates to
* {@link #spawn(SpawnRequest)} with the decision's profile named explicitly — its only spawn
* contract, since there is no separate routing path to honor.
* <p>No default implementation (fleetd #450): the two correct bodies disagree on purpose, so an
* implementer must choose one rather than silently inherit whichever this interface happened to
* provide. An implementer with no placement concept of its own — spawns a single profile, e.g.
* {@code HerdrPeerLauncher} — should delegate to {@link #spawn(SpawnRequest)} with the decision's
* profile named explicitly, since there is no separate routing path to honor there: the
* explicit-profile branch it re-enters and the routing branch {@link #place} would have used are
* the same thing. <strong>A launcher that routes across more than one profile — the way {@code
* CompositePeerLauncher} routes across every configured adapter — MUST NOT re-enter {@link
* #spawn(SpawnRequest)}.</strong> Doing so re-applies that single-argument method's
* explicit-profile checks ({@code enforceNotQuarantined}, {@code enforceNotCoolingOff}, {@code
* enforceMaxLoad}, {@code enforceModelEnabled} in {@code CompositePeerLauncher}), which can
* refuse the very profile {@link #place} just chose, if the underlying placement state moved in
* the window between the {@link #place} call and this one — the exact window this method and
* {@link PlacementDecision} exist to close (fleetd #444). Before #450 this was a {@code default}
* method that only {@code CompositePeerLauncher} overrode; a future placement-doing launcher
* could have inherited the re-entering body silently and never known. Making it abstract turns
* that silent inheritance into a compile error.
*
* @throws IllegalArgumentException if the decision names an unknown profile
*/
default PeerHandle spawn(SpawnRequest req, PlacementDecision decision) {
return spawn(req.withProfile(decision.profile()));
}
PeerHandle spawn(SpawnRequest req, PlacementDecision decision);
/**
* Resolve the effective working directory for a spawn {@code req} without actually spawning.
@@ -20,6 +20,7 @@ import dev.ltms.fleet.peer.PeerLauncher;
import dev.ltms.fleet.peer.SpawnRequest;
import dev.ltms.fleet.placement.BackendOutagePolicy;
import dev.ltms.fleet.placement.BackendQuarantine;
import dev.ltms.fleet.placement.PlacementDecision;
import dev.ltms.fleet.placement.PlacementPolicies;
import dev.ltms.fleet.session.MemberSession;
import dev.ltms.fleet.session.SessionManager;
@@ -123,6 +124,11 @@ class FleetdBackendErrorSinkTest {
throw new UnsupportedOperationException("not reachable — this test never acquires a session");
}
@Override
public PeerHandle spawn(SpawnRequest req, PlacementDecision decision) {
throw new UnsupportedOperationException("not reachable — this test never acquires a session");
}
@Override
public Set<String> profiles() {
return Set.of();
@@ -20,6 +20,7 @@ import dev.ltms.fleet.peer.PeerUnreachableException;
import dev.ltms.fleet.peer.SpawnRequest;
import dev.ltms.fleet.placement.BackendOutagePolicy;
import dev.ltms.fleet.placement.BackendQuarantine;
import dev.ltms.fleet.placement.PlacementDecision;
import dev.ltms.fleet.placement.PlacementException;
import dev.ltms.fleet.placement.PlacementPolicies;
import org.junit.jupiter.api.Test;
@@ -1123,6 +1124,68 @@ class CompositePeerLauncherTest {
assertEquals(0, adapter.spawnCount("b"), "routedProfileFor never spawns anything");
}
/**
* fleetd #444: {@link PlacementDecision} exists to close the window between {@link
* CompositePeerLauncher#place} and {@link CompositePeerLauncher#spawn(SpawnRequest,
* PlacementDecision)} — the placement state must be free to move in that window without the
* held decision being re-checked against the new state. Every quarantine test above resolves
* and spawns in one call, so none of them ever open that window; this test is the one that
* does: "sol" is placed FIRST, while nothing is quarantined yet, and only THEN is its
* credential quarantined, before the held decision is spawned.
*
* <p>This is the test that tells the real override apart from the alternative body the ticket
* measured: routing {@code decision.profile()} straight to its adapter (the real override)
* never re-runs {@code enforceNotQuarantined}, so the spawn against the held decision still
* succeeds on sol. Re-entering {@code spawn(req.withProfile(decision.profile()))} instead
* lands in the explicit-profile branch, which refuses a now-quarantined sol outright — before
* this test existed, replacing the real override's body with that re-entering call left the
* whole suite green.
*/
@Test
void spawnHonorsAPlacementDecisionEvenAfterItsProfileIsQuarantinedInTheWindowAfterPlace() {
FakeHerdr herdr = new FakeHerdr();
Map<String, FleetConfig.Profile> profiles = ordered(
"sol", stubWorker("sol", "shared-openai"),
"b", stubWorker("b"));
// The adapter's OWN fallback default is "b", deliberately different from the profile place()
// decides ("sol") — see the note below on why this must not be "sol" too.
StubLauncher adapter = new StubLauncher("claude", herdr, profiles, "b", Set.of());
BackendQuarantine quarantine = new BackendQuarantine(() -> 0L, TimeUnit.MINUTES.toNanos(30));
CompositePeerLauncher composite = new CompositePeerLauncher(List.of(adapter), "sol", profiles,
PlacementPolicies.fixed(), _ -> 0, null, quarantine);
// 1. Resolve BEFORE anything is quarantined — sol (definition order first, fixed policy) wins.
// composite's own defaultProfile ("sol", the constructor arg above) never enters this: the
// pool poolFor(DEV) resolves to is never empty here, so place() only ever reads that field as
// a fallback for an empty pool, which this test does not exercise.
PlacementDecision decision = composite.place(MemberRole.DEV);
assertEquals("sol", decision.profile(), "sanity: nothing is quarantined yet, so sol is placed");
// 2. Move the placement state IN THE WINDOW between place() and spawn() — sol's credential
// is now quarantined. A fresh place()/spawn(req) pair would fall through to b instead; the
// held decision must not be re-evaluated against this new state at all.
quarantine.quarantine("shared-openai");
// 3. Spawn against the HELD decision, not a fresh resolve.
SpawnRequest req = new SpawnRequest(null, null, null, null, null, MemberRole.DEV);
PeerHandle handle = composite.spawn(req, decision);
assertEquals("sol", handle.profile(),
"the decision from place() is honored even though sol is now quarantined");
// A fixture whose adapter falls back to "sol" too would let an UNSTAMPED request (one
// routed but never given req.withProfile("sol")) land on spawnCount("sol") == 1 by
// COINCIDENCE, since StubLauncher.spawn falls back to its own defaultProfile whenever
// req.profileName() is blank. Giving the adapter "b" as its fallback instead means only an
// actually-stamped request can produce this count — an unstamped one would count against
// "b" and this assertion would fail.
assertEquals(1, adapter.spawnCount("sol"),
"the request that reached the delegate actually carried sol as its profile "
+ "(the adapter's own fallback default is 'b', so this can't happen by accident)");
assertEquals(0, adapter.spawnCount("b"),
"b must never be touched — neither as the decision's profile nor as an unstamped "
+ "request's accidental fallback");
}
@Test
void aQuarantineLiftsOnTheInjectedClockAndTheProfileBecomesSpawnableAgain() {
FakeHerdr herdr = new FakeHerdr();
@@ -23,6 +23,7 @@ import dev.ltms.fleet.peer.PeerLauncher;
import dev.ltms.fleet.peer.PeerUnreachableException;
import dev.ltms.fleet.peer.SpawnRequest;
import dev.ltms.fleet.placement.BackendQuarantine;
import dev.ltms.fleet.placement.PlacementDecision;
import dev.ltms.fleet.placement.PlacementPolicies;
import org.junit.jupiter.api.Test;
import org.junit.jupiter.api.io.TempDir;
@@ -1036,6 +1037,11 @@ class SessionManagerTest {
return delegate.spawn(req);
}
@Override
public PeerHandle spawn(SpawnRequest req, PlacementDecision decision) {
return delegate.spawn(req, decision);
}
@Override
public Set<String> profiles() {
return delegate.profiles();
@@ -1597,6 +1603,11 @@ class SessionManagerTest {
throw new UnsupportedOperationException("not reachable — the capability check refuses first");
}
@Override
public PeerHandle spawn(SpawnRequest req, PlacementDecision decision) {
throw new UnsupportedOperationException("not reachable — the capability check refuses first");
}
@Override
public Set<String> profiles() {
return Set.of("stub-profile");
@@ -1661,6 +1672,11 @@ class SessionManagerTest {
return delegate.spawn(req);
}
@Override
public PeerHandle spawn(SpawnRequest req, PlacementDecision decision) {
return delegate.spawn(req, decision);
}
@Override
public Set<String> profiles() {
return delegate.profiles();
@@ -1867,6 +1883,11 @@ class SessionManagerTest {
return handle;
}
@Override
public PeerHandle spawn(SpawnRequest req, PlacementDecision decision) {
return spawn(req.withProfile(decision.profile()));
}
@Override
public Set<String> profiles() {
return Set.of("lazy");