refactor(plan): migrate Acquirer.Acquire to lagotto/pkg/snipe.Snipe (lagotto#106)

Replace the hand-rolled retry/backoff/AZ-sweep/classify loop in
Acquirer.Acquire with a thin wrapper over snipe.Snipe, now that lagotto
v0.52.1 ships it with region-pinned client resolution (lagotto#111).
Deletes ~180 lines of duplicated logic; SpawnLauncher becomes a pure
LaunchConfig builder (Client/Timeout/Provision removed).

Local retry-behavior tests are dropped in favor of lagotto/pkg/snipe's
own 16-test suite. The Substrate-backed offline test tier for Acquirer
is also lost, since Snipe builds its own client internally with no
custom-endpoint injection point — filed lagotto#113 to restore it.

Verified live against real AWS with Placements populated (calque's
actual usage pattern); also found and filed lagotto#114 (AZ field
bug), which doesn't affect calque since all call sites pin Placements.

Author: scttfrdmn

cmd/calque/fleetrun.go

@@ -178,12 +178,12 @@ func runShard(ctx context.Context, s3c *s3.Client, spawnClient *spawnaws.Client,
 		ManifestKey: shardLayout.ManifestKey, WorkerDir: hostWorkerDir, Region: o.region,
 		LogKey: shardLayout.LogKey, HostMode: false, ModelEnv: o.model,
 	}
-	launcher := &plan.SpawnLauncher{
-		Client: spawnClient, RunCmd: boot.Command(), TTL: o.ttl, OnComplete: "terminate",
-		Username: "ubuntu", Timeout: 5 * time.Minute, AMI: o.ami, PricePerHour: pricePerHr,
+	launchCfg := plan.SpawnLauncher{
+		RunCmd: boot.Command(), TTL: o.ttl, OnComplete: "terminate",
+		Username: "ubuntu", AMI: o.ami, PricePerHour: pricePerHr,
 		IMDSv2HopLimit: 2, RootVolumeGiB: 200,
-	}
-	acq := &plan.Acquirer{Launcher: launcher, Report: rep.rep, Deadline: o.deadline, Placements: places}
+	}.Build()
+	acq := &plan.Acquirer{LaunchConfig: launchCfg, Report: rep.rep, Deadline: o.deadline, Placements: places}
 	tgt := &target.Target{Card: target.DefaultCard, Instance: o.instance}
 	acquired, err := acq.Acquire(ctx, tgt, o.region)
 	if err != nil {

cmd/calque/realrun.go

@@ -135,14 +135,14 @@ func realRun(o realOpts) (err error) {
 	if err != nil {
 		return fmt.Errorf("spawn client: %w", err)
 	}
-	launcher := &plan.SpawnLauncher{
-		Client: spawnClient, RunCmd: boot.Command(), TTL: o.ttl, OnComplete: "terminate",
-		Username: "ubuntu", Timeout: 5 * time.Minute, AMI: o.ami, PricePerHour: pricePerHr,
+	launchCfg := plan.SpawnLauncher{
+		RunCmd: boot.Command(), TTL: o.ttl, OnComplete: "terminate",
+		Username: "ubuntu", AMI: o.ami, PricePerHour: pricePerHr,
 		IMDSv2HopLimit: 2,   // warmd runs INSIDE docker; needs IMDS creds one hop away
 		RootVolumeGiB:  200, // vLLM image + weights blow past spawn's 20 GiB default
-	}
+	}.Build()
 	acq := &plan.Acquirer{
-		Launcher: launcher, Report: rep, Deadline: o.deadline, Placements: places,
+		LaunchConfig: launchCfg, Report: rep, Deadline: o.deadline, Placements: places,
 		OnProgress: func(attempt int, code, detail string, waited time.Duration) {
 			fmt.Printf("      ...swept %d attempt(s), no capacity (%s, %s)\n", attempt, code, waited.Round(time.Second))
 		},

cmd/calque/session.go

@@ -107,13 +107,13 @@ func runSession(o sessionOpts) (err error) {
 		}
 	}
 
-	launcher := &plan.SpawnLauncher{
-		Client: spawnClient, RunCmd: prep.PrepCommand(artifactPfx), TTL: o.ttl,
+	launchCfg := plan.SpawnLauncher{
+		RunCmd: prep.PrepCommand(artifactPfx), TTL: o.ttl,
 		OnComplete: "", // do NOT terminate on command completion — we hold the box
-		Username:   "ubuntu", Timeout: 5 * time.Minute, AMI: o.ami, PricePerHour: pricePerHr,
+		Username:   "ubuntu", AMI: o.ami, PricePerHour: pricePerHr,
 		IMDSv2HopLimit: 2, RootVolumeGiB: 200,
 		Spot: o.spot, SpotMaxPrice: o.spotMaxPrice,
-	}
+	}.Build()
 	if o.spot {
 		// Honesty (§9/§10): a spot ramp measures K against a SPOT R_a, and the box
 		// can be reclaimed mid-ramp. Say so loudly and leak it, so the resulting K
@@ -130,7 +130,7 @@ func runSession(o sessionOpts) (err error) {
 	round := 0
 	lastDetail := ""
 	acq := &plan.Acquirer{
-		Launcher: launcher, Report: rep, Deadline: o.acquireDeadline, Placements: places,
+		LaunchConfig: launchCfg, Report: rep, Deadline: o.acquireDeadline, Placements: places,
 		OnProgress: func(attempt int, code, detail string, waited time.Duration) {
 			// Print the full AWS message on the first round, whenever it CHANGES, and
 			// every 10th round — so a capacity opening (message names an AZ) or a

cmd/calque/smoke.go

@@ -121,12 +121,12 @@ func smoke(o smokeOpts) (err error) {
 	if err != nil {
 		return fmt.Errorf("spawn client: %w", err)
 	}
-	launcher := &plan.SpawnLauncher{
-		Client: spawnClient, RunCmd: boot.Command(), TTL: o.ttl, OnComplete: "terminate",
-		Username: "ubuntu", Timeout: 5 * time.Minute, AMI: o.ami, PricePerHour: pricePerHr,
-	}
+	launchCfg := plan.SpawnLauncher{
+		RunCmd: boot.Command(), TTL: o.ttl, OnComplete: "terminate",
+		Username: "ubuntu", AMI: o.ami, PricePerHour: pricePerHr,
+	}.Build()
 	acq := &plan.Acquirer{
-		Launcher: launcher, Report: rep, Deadline: o.deadline, Placements: places,
+		LaunchConfig: launchCfg, Report: rep, Deadline: o.deadline, Placements: places,
 		OnProgress: func(attempt int, code, detail string, waited time.Duration) {
 			fmt.Printf("      ...swept %d attempt(s), still no capacity (%s, %s)\n", attempt, code, waited.Round(time.Second))
 		},

cmd/calque/spawnrun.go

@@ -175,11 +175,11 @@ func runSpawnShard(ctx context.Context, s3c *s3.Client, spawnClient *spawnaws.Cl
 		Bucket: o.bucket, ArtifactPrefix: shardLayout.ArtifactPfx, ManifestKey: shardLayout.ManifestKey,
 		WorkerDir: hostWorkerDir, Region: o.region, LogKey: shardLayout.LogKey, HostMode: true,
 	}
-	launcher := &plan.SpawnLauncher{
-		Client: spawnClient, RunCmd: boot.Command(), TTL: o.ttl, OnComplete: "terminate",
-		Username: "ubuntu", Timeout: 5 * time.Minute, AMI: o.ami,
-	}
-	acq := &plan.Acquirer{Launcher: launcher, Report: rep.rep, Deadline: o.deadline, Placements: places}
+	launchCfg := plan.SpawnLauncher{
+		RunCmd: boot.Command(), TTL: o.ttl, OnComplete: "terminate",
+		Username: "ubuntu", AMI: o.ami,
+	}.Build()
+	acq := &plan.Acquirer{LaunchConfig: launchCfg, Report: rep.rep, Deadline: o.deadline, Placements: places}
 	tgt := &target.Target{Card: target.DefaultCard, Instance: o.instance}
 	acquired, err := acq.Acquire(ctx, tgt, o.region)
 	if err != nil {

cmd/gpuprobe/main.go

@@ -111,14 +111,14 @@ func probe(instanceType, region, ami, ttl string, deadline time.Duration, spot b
 		}
 	}
 
-	launcher := &plan.SpawnLauncher{
-		Client: spawnClient, TTL: ttl, OnComplete: "terminate",
-		Username: "ubuntu", Timeout: 5 * time.Minute, AMI: ami, Spot: spot,
+	launchCfg := plan.SpawnLauncher{
+		TTL: ttl, OnComplete: "terminate",
+		Username: "ubuntu", AMI: ami, Spot: spot,
 		// No RunCmd: boot plain, no docker/GPU job — this probe runs over SSM
 		// after the instance is up, not via user-data.
-	}
+	}.Build()
 	acq := &plan.Acquirer{
-		Launcher: launcher, Report: rep, Deadline: deadline, Placements: places,
+		LaunchConfig: launchCfg, Report: rep, Deadline: deadline, Placements: places,
 		OnProgress: func(attempt int, code, detail string, waited time.Duration) {
 			fmt.Printf("      ...swept %d attempt(s), no capacity (%s, %s)\n", attempt, code, waited.Round(time.Second))
 		},

docs/substrate-offline-test-tier.md

@@ -1,15 +1,45 @@
 # Design note: Substrate as calque's offline AWS test tier (calque#114)
 
-**Status:** shipped, with one documented scope reduction from the original
-issue text. `github.com/scttfrdmn/substrate/emulator` (Scott's event-sourced
-AWS emulator) is now calque's middle test tier: hand-written fakes (today's
-`plan_test.go` `fakeResolver`/`scriptedLauncher` pattern, unchanged) at one
-end, real-`AWS_PROFILE=aws` spend at the other, and Substrate in between —
-real request/response wire round-trips, no billing.
+**Status:** shipped, then PARTIALLY REGRESSED by calque#106's later
+`Acquirer` → `lagotto/pkg/snipe.Snipe` migration — see "Regression" below.
+`github.com/scttfrdmn/substrate/emulator` (Scott's event-sourced AWS
+emulator) was calque's middle test tier: hand-written fakes at one end,
+real-`AWS_PROFILE=aws` spend at the other, Substrate in between — real
+request/response wire round-trips, no billing. That middle tier for
+`Acquirer` specifically no longer exists as of the `Snipe` migration; see
+below for why and what's tracked to restore it.
 
-## What shipped
+## Regression: the Acquirer↔Substrate tests were deleted (calque#106 migration)
 
-`internal/plan/substrate_test.go`:
+`internal/plan/substrate_test.go` (originally `TestAcquireAgainstSubstrate`/
+`TestAcquireAgainstSubstrateInjectedFailure`, both described below in their
+original, now-historical form) was DELETED when `Acquirer.Acquire` was
+migrated to delegate to `lagotto/pkg/snipe.Snipe` (lagotto#106/#111,
+calque commit history — the same migration that deleted calque's own
+hand-rolled retry/backoff/AZ-sweep/classify loop and its 5 local tests,
+per `internal/plan/plan_test.go`'s own updated comments).
+
+The reason: `Snipe` builds its own `*spawnaws.Client` internally via
+`spawnaws.NewClientWithRegion` (region-pinned as of lagotto#111), which
+loads AWS config through the default credential chain with NO way to
+point at a custom endpoint. The only constructor that supports a custom
+endpoint, `spawnaws.NewClientFromConfig`, is unreachable from `Snipe` —
+its internal `clientFor` is unexported and `Options`/`Target` have no
+field for injecting a client or a custom `aws.Config`. Filed upstream:
+[lagotto#113](https://github.com/spore-host/lagotto/issues/113).
+
+Until that lands, `Acquirer`'s real request-building/response-parsing
+code has NO offline test tier at all — only real `AWS_PROFILE=aws` runs
+exercise it (as they already did for the retry/backoff logic before
+`Snipe`, and as `fleetrun.go`'s own historical untested-without-spend
+precedent already established for this exact code path). This is a real,
+acknowledged loss of coverage, accepted as the cost of deleting ~180 lines
+of hand-rolled retry logic in favor of a well-tested (16 tests) upstream
+leaf — re-evaluate once lagotto#113 lands.
+
+## What shipped (historical — describes the NOW-DELETED tests, kept for context)
+
+`internal/plan/substrate_test.go` (deleted, see above):
 
 - **`TestAcquireAgainstSubstrate`** — points the REAL `plan.SpawnLauncher`
   (wrapping spawn's real `*spawnaws.Client`/`launcher.Provision`) at a
@@ -26,7 +56,7 @@ real request/response wire round-trips, no billing.
   is handled by the bounded-retry-then-fail-fast path — never an infinite
   loop on an unclassifiable error.
 
-Both tests run with `PricePerHour` pinned on the launcher (skips spawn's own
+Both tests ran with `PricePerHour` pinned on the launcher (skips spawn's own
 live AWS Pricing API call — a real network dependency separate from EC2 that
 would otherwise sneak into an "offline" test) and a pinned `AMI` (skips AMI
 auto-detection's SSM round-trip, which Substrate would need separately

go.mod

@@ -14,7 +14,7 @@ require (
 	github.com/aws/smithy-go v1.27.3
 	github.com/scttfrdmn/substrate v0.94.0
 	github.com/spore-host/cohort v0.2.0
-	github.com/spore-host/lagotto v0.52.0
+	github.com/spore-host/lagotto v0.52.1
 	github.com/spore-host/spawn v0.98.0
 	github.com/spore-host/truffle v0.48.1
 )

go.sum

@@ -405,8 +405,8 @@ github.com/spf13/viper v1.21.0 h1:x5S+0EU27Lbphp4UKm1C+1oQO+rKx36vfCoaVebLFSU=
 github.com/spf13/viper v1.21.0/go.mod h1:P0lhsswPGWD/1lZJ9ny3fYnVqxiegrlNrEmgLjbTCAY=
 github.com/spore-host/cohort v0.2.0 h1:PGKkYawNzymE6jPebRqsZHK9X/m2UVYupujMnIdK8f4=
 github.com/spore-host/cohort v0.2.0/go.mod h1:sNMWDccvNp3Qso8ZVMcOvJTJ8GC/QY9LDUYQK1zYWLc=
-github.com/spore-host/lagotto v0.52.0 h1:UMl0ZnbFauPPhZDXRo903HrwUAnpUa96XpHOtx82Rzg=
-github.com/spore-host/lagotto v0.52.0/go.mod h1:ljK6zEq0cd+sxsIksJo4Y6E8eHji756exsIquNmNrao=
+github.com/spore-host/lagotto v0.52.1 h1:rqKN2VvNmGk+gxYfS9FeaqL4ROZ70vMWQkxiOiZnc2w=
+github.com/spore-host/lagotto v0.52.1/go.mod h1:ljK6zEq0cd+sxsIksJo4Y6E8eHji756exsIquNmNrao=
 github.com/spore-host/libs v0.43.3 h1:Pa/DC49S8uxkmhodyi7x2nnu19qZ2uqmwZEnb6ATi1k=
 github.com/spore-host/libs v0.43.3/go.mod h1:q9UOt1DiO8Zmz4t5EWbSQV+r9FQsUOYaRkCK6TE60G4=
 github.com/spore-host/spawn v0.98.0 h1:Yp05t9UtjJBLNdwWdcuB+6sVfb9VsYWATb4oPdTKuiQ=