@evident-ai/runner-cdk 3.4.1-dev.2f1b44b → 3.4.1-dev.38181ff

package/dist/microvm-image-context/hooks/resume

@@ -1,14 +1,9 @@
 #!/usr/bin/env bash
 #
 # Re-dial with the identity /run left behind. There is no step that could fetch
-# a fresh runner key, so the one from /run is what resumes. Also restarts the
-# session-DB replicator /suspend stopped before the snapshot (#812 WI-4):
-# litestream does not survive a suspend/resume freeze on this design (Q2) —
-# /suspend stops it and /resume starts a fresh one, the same pattern already
-# proven for the tunnel. The interval credential sync loop (#1868 WI-3) is the
-# same story one function over: /suspend stops it too, so a resumed VM that
-# never restarted it here would never sync credentials again for the rest of
-# its life.
+# a fresh runner key, so the one from /run is what resumes. The CLI delegated by
+# start_tunnel owns the credential loop and the session-DB replicator's
+# post-resume start.
 set -euo pipefail
 
 # shellcheck source=./common.sh
@@ -42,14 +37,18 @@ fi
   read -r runner_key
 } <"${CONTEXT_FILE}"
 
+# Restore the id so the resumed CLI can self-report and acknowledge any
+# outstanding recycle request (#1906). Older images may not have this file.
+if [ -s "${MICROVM_ID_FILE}" ]; then
+  MICROVM_ID="$(cat "${MICROVM_ID_FILE}")"
+  export MICROVM_ID
+fi
+
 # Tolerant, never fatal: a resume that fails costs the user their whole
 # session, so a broken durable-state config degrades to "no session-DB
-# restart" rather than a failed resume (the same Q3 reasoning /run's
-# session-DB step follows). start_litestream's own guards (disabled/marker/
-# already-running) handle the rest — this needs no logic of its own.
+# replication" rather than a failed resume. The CLI's own guards handle the
+# rest — this needs no logic of its own.
 load_state_config || warn "could not resolve durable-state config; the session DB will not resume replicating"
-start_litestream
-start_credential_sync
 
 # Diagnostic-only, unlike /run's gate: a failed resume costs the user their
 # whole session, so this never exits — it only converts a silent "resumed but
@@ -74,9 +73,8 @@ start_credential_sync
 # list, so a later edit cannot accidentally foreground the check again.
 ( check_runner_key "${runner_key}" "${api_url}" || true ) &
 
-# Not waited on: the CLI backs off and keeps retrying its own dial with no
-# attempt cap (apps/cli/src/lib/tunnel/connection.ts), and a control-plane
-# reclaim pass (the api-worker lifecycle cron) is what decides whether a
+# Not waited on: the Evident CLI backs off and keeps retrying its own dial with no
+# attempt cap, and Evident's control-plane reclaim pass is what decides whether a
 # runner that never reconnects gets suspended.
 start_tunnel "${runner_key}" "${api_url}" "${tunnel_url}"
 

package/dist/microvm-image-context/hooks/run

@@ -28,7 +28,6 @@ cleanup() {
   stop_tunnel || warn "stop_tunnel failed while cleaning up"
   stop_opencode || warn "stop_opencode failed while cleaning up"
   kill_litestream || warn "kill_litestream failed while cleaning up"
-  stop_credential_sync || warn "stop_credential_sync failed while cleaning up"
 }
 trap cleanup EXIT
 
@@ -52,20 +51,20 @@ regenerate_machine_id
 # reads it, and left behind for the hooks that flush back to the same prefix.
 printf '%s\n' "${state_prefix}" >"${STATE_PREFIX_FILE}"
 
-# 3 — start reading the litestream binary NOW so that read overlaps step 4,
-# which needs a different binary (node), instead of landing inside step 5, which
-# start_opencode has to wait for. Backgrounded and never waited on.
+# 3 — start reading the litestream and aws CLI binaries NOW. The reads run in
+# the background and overlap the CLI's own node startup and auth round trip,
+# keeping their cold first touches off the path to the first credential fetch.
 prewarm_litestream
+prewarm_aws_cli
 
-# 4 — credential stores, restored before anything that reads them starts. Fatal
-# only if persistence itself is unavailable (an unset bucket/prefix, or a broken
-# synchroniser bundle) — a VM with no model credentials yet still boots.
-restore_credentials
+# 4 — export the durable-state config before anything that reads it. Fatal,
+# because the CLI and the teardown hooks must agree on this runner's state prefix.
+load_state_config || exit 1
 
 # 5 — does the runner key in this payload actually authenticate against
-# Evident? Checked here, before opencode or the tunnel spend any of this boot's
-# SIGTERM budget on a key that cannot work, and before anything has started
-# that cleanup would need to tear down. Fatal ONLY on contrary evidence
+# Evident? Checked here, before the tunnel spends any of this boot's time on a key
+# that cannot work, and before anything has started that cleanup would need to
+# tear down. Fatal ONLY on contrary evidence
 # (check_runner_key's own contract, common.sh): a key the API actively rejects
 # cannot work regardless, and failing here costs ~5s against the ~10 minutes a
 # runner that can never connect would otherwise burn before the lifecycle cron
@@ -73,49 +72,14 @@ restore_credentials
 # it warns and this VM still boots.
 check_runner_key "${runner_key}" "${api_url}" || exit 1
 
-# 6 — the OpenCode session DB (#812 WI-2), restored before opencode opens it —
-# the only place this can happen. Bare, like restore_credentials above:
-# restore_session_db never returns non-zero (Q3 — nothing about the session
-# DB may ever fail /run), so there is deliberately no `||` here to catch.
-restore_session_db
-log "session DB restore done ${SECONDS}s into the hook"
-
-# 7 — integrity-check the restored DB (#1868 WI-4), after the config file
-# exists (verify_session_db reads it) and before anything opens the DB —
-# the only window in which that's true. The ONE step in this sequence that
-# can still fail /run past the runner-key gate: a 34 means the corrupt DB's
-# separation/disposal could not be proven safe (see verify_session_db's own
-# comment for why that's a deliberate exception to "nothing about the
-# session DB may ever fail /run").
-verify_session_db || exit 1
-
-# 8 — apply the overlay before OpenCode resolves its project configuration.
-apply_runner_opencode_config
-
-# 9 — configure git after credentials are restored and before agent shells start.
-configure_github_access
-
-# 10 — opencode. Started here, not at build time: a warm process in the shared
-# snapshot would carry its installation id and database into every VM. Not
-# waited on: a slow opencode boot is not a reason to fail /run (the tunnel CLI
-# auto-starts opencode when it finds none healthy,
-# apps/cli/src/commands/ensure-opencode.ts, and the api-worker lifecycle cron
-# reclaims a runner that never comes online).
-start_opencode
-
-# 11 — begin replicating the session DB (#812 WI-3), now that opencode has
-# opened it and before any work can arrive over the tunnel. Bare, like
-# restore_session_db above: start_litestream never returns non-zero (every
-# guard inside it is its own `return 0`), so there is nothing here for
-# `set -e` to abort on.
-start_litestream
-
-# 11a — the interval credential sync (#1868 WI-3), matching ECS's own
-# post-litestream position. Bare for the same reason: every guard inside
-# start_credential_sync is its own `return 0`.
-start_credential_sync
+# 6 — preserve the VM id across suspend/resume. The runtime always supplies it
+# for a /run hook, and the subshell keeps the persisted value protected.
+(
+  umask 077
+  printf '%s\n' "${MICROVM_ID}" >"${MICROVM_ID_FILE}"
+)
 
-# 12 — the first per-VM identity on the wire. The subshell's umask makes the file
+# 7 — the first per-VM identity on the wire. The subshell's umask makes the file
 # unreadable to anyone else from the moment it exists, before the key is in it.
 (
   umask 077
@@ -123,10 +87,9 @@ start_credential_sync
     >"${CONTEXT_FILE}"
 )
 
-# Not waited on: the CLI backs off and keeps retrying its own dial with no
-# attempt cap (apps/cli/src/lib/tunnel/connection.ts), and a control-plane
-# reclaim pass (the api-worker lifecycle cron) is what decides whether a
+# Not waited on: the Evident CLI backs off and keeps retrying its own dial with no
+# attempt cap, and Evident's control-plane reclaim pass is what decides whether a
 # runner that never connects gets suspended.
-start_tunnel "${runner_key}" "${api_url}" "${tunnel_url}"
+start_tunnel "${runner_key}" "${api_url}" "${tunnel_url}" true true
 
 run_succeeded=true

package/dist/microvm-image-context/hooks/suspend

@@ -8,11 +8,10 @@ set -euo pipefail
 # shellcheck source=./common.sh
 source "$(dirname "$0")/common.sh"
 
-# Stopped BEFORE the boundary flush (#1868 WI-3): the interval loop and this
-# flush must not race each other on the same credential stores.
-stop_credential_sync
-sync_credentials
-stop_tunnel
+# load_state_config exports PERSISTENCE_BUCKET, which flush_session_db below
+# requires; the CLI flush completes, or is proven absent, before teardown continues.
+load_state_config || warn "could not resolve durable-state config; neither the credential flush nor the session-DB flush below can run"
+stop_runner_and_flush_credentials
 
 # opencode is deliberately NOT stopped here (#812 WI-4) — it is inside the
 # snapshot and must resume with it — so a turn still in flight may not be

package/dist/microvm-image-context/hooks/terminate

@@ -8,11 +8,10 @@ set -uo pipefail
 # shellcheck source=./common.sh
 source "$(dirname "$0")/common.sh"
 
-# Stopped BEFORE the boundary flush (#1868 WI-3): the interval loop and this
-# flush must not race each other on the same credential stores.
-stop_credential_sync
-sync_credentials
-stop_tunnel
+# load_state_config exports PERSISTENCE_BUCKET, which flush_session_db below
+# requires; the CLI flush completes, or is proven absent, before teardown continues.
+load_state_config || warn "could not resolve durable-state config; neither the credential flush nor the session-DB flush below can run"
+stop_runner_and_flush_credentials
 
 # Writers stopped BEFORE litestream's final sync (#812 WI-4, the
 # ordered-shutdown invariant, plan §6): otherwise litestream could snapshot
@@ -32,6 +31,6 @@ flush_session_db
 # this hook runs under `set -u` (above), so an unbound one aborts the script
 # AT THIS LINE, leaving the runner key sitting in CONTEXT_FILE on a VM that is
 # being torn down. That is the one thing this line exists to prevent.
-rm -f "${CONTEXT_FILE}" "${SESSION_DB_NO_REPLICATE_MARKER}"
+rm -f "${CONTEXT_FILE}" "${SESSION_DB_NO_REPLICATE_MARKER}" "${CREDENTIAL_FLUSH_MARKER_FILE}"
 
 exit 0

package/dist/waker/construct.js

@@ -60,8 +60,7 @@ class EvidentWaker extends constructs_1.Construct {
             // dist/waker-lambda/handler.js, so neither this package's published npm
             // artifact (which ships dist/ already built) nor a consuming app needs
             // that private workspace package, or esbuild, on synth's PATH. Requires
-            // a build before synth when consuming this package via `workspace:*` —
-            // see infrastructure/evident-runner's README "local gate".
+            // a build before synth when consuming this package via `workspace:*`.
             //
             // Resolved from the PACKAGE ROOT (../.. ), not `__dirname` directly:
             // `__dirname` is `dist/waker` at runtime (compiled) but `src/waker` when

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@evident-ai/runner-cdk",
-  "version": "3.4.1-dev.2f1b44b",
+  "version": "3.4.1-dev.38181ff",
   "description": "Reusable CDK constructs for an Evident agent runner: a single scale-to-zero Fargate runner (task + service + per-agent self-stop role + waker Lambda), or a per-session AWS Lambda MicroVM that boots on demand and suspends between messages. Instantiate once per agent from your own stack.",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",
@@ -14,8 +14,8 @@
     "dist"
   ],
   "scripts": {
-    "build": "pnpm run build-bundled-deps && ts-node scripts/build.ts",
-    "//build-bundled-deps": "scripts/build.ts esbuild-bundles runner/docker-images/microvm/hook-server.ts, which imports @evident-ai/lambda-microvm-runtime by its built `main`. @evident-ai/lambda-microvm-cdk is a workspace:* devDependency whose types resolve through its own dist/, so tsc --project tsconfig.build.json cannot compile src/microvm/{shapes,construct}.ts without it. Chained into `build` rather than left to turbo's `^build` because publish-runner-cdk.yaml and infrastructure/evident-runner's build-runner-cdk-dep invoke `pnpm --filter @evident-ai/runner-cdk build` directly, which bypasses turbo entirely.",
+    "build": "pnpm run build-bundled-deps && ts-node scripts/build.ts && node -e 'const m=require(\"./package.json\").main; if (!require(\"node:fs\").existsSync(m)) { console.error(`build produced no ${m}`); process.exit(1); }'",
+    "//build-bundled-deps": "scripts/build.ts esbuild-bundles runner/docker-images/microvm/hook-server.ts, which imports @evident-ai/lambda-microvm-runtime by its built `main`. @evident-ai/lambda-microvm-cdk is a workspace:* devDependency whose types resolve through its own dist/, so tsc --project tsconfig.build.json cannot compile src/microvm/{shapes,construct}.ts without it. Chained into `build` rather than left to turbo's `^build` because callers outside turbo invoke `pnpm --filter @evident-ai/runner-cdk build` directly, which bypasses turbo entirely.",
     "build-bundled-deps": "pnpm --filter @evident-ai/lambda-microvm-cdk build && pnpm --filter @evident-ai/lambda-microvm-runtime build",
     "typecheck": "tsc --noEmit",
     "test": "node --test --require ts-node/register 'src/**/*.test.ts'",