@go-to-k/cdkd 0.284.52 → 0.284.54

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -1,5 +1,5 @@
1
1
  import { a as getLiveRenderer, d as generateResourceNameWithFallback, f as getCurrentStackName, h as withStackName, l as applyDefaultNameForFallback, n as getLogger, p as looksLikeCdkdGeneratedName, u as generateResourceName } from "./logger-zRrlbaQt.js";
2
- import { t as getCdkdVersion } from "./version-BXmUXkY2.js";
2
+ import { t as getCdkdVersion } from "./version-DLRgymoe.js";
3
3
  import { randomUUID } from "node:crypto";
4
4
  import { CreateBucketCommand, DeleteObjectCommand, DeleteObjectsCommand, GetBucketLocationCommand, GetObjectCommand, HeadBucketCommand, HeadObjectCommand, ListObjectsV2Command, NoSuchKey, PutBucketEncryptionCommand, PutBucketPolicyCommand, PutObjectCommand, PutPublicAccessBlockCommand, S3Client, S3ServiceException } from "@aws-sdk/client-s3";
5
5
  import { CloudControlClient, CreateResourceCommand, DeleteResourceCommand, GetResourceCommand, GetResourceRequestStatusCommand, ListResourcesCommand, UpdateResourceCommand } from "@aws-sdk/client-cloudcontrol";
@@ -110,11 +110,17 @@ const IAM_PROPAGATION_ERROR_MESSAGE_PATTERNS = [
110
110
  "is not a valid role to allow SNS"
111
111
  ];
112
112
  /**
113
- * The NON-IAM-propagation half of {@link RETRYABLE_ERROR_MESSAGE_PATTERNS}:
114
- * transient failures whose recovery window is either long (SQS's 60s same-name
115
- * cooldown, a resource still leaving a Pending/Creating state) or genuinely
116
- * load-related (throttling), where hammering AWS with dense retries is harmful
117
- * and exponential backoff is the correct shape.
113
+ * The NON-IAM-propagation, NON-name-cooldown third of
114
+ * {@link RETRYABLE_ERROR_MESSAGE_PATTERNS}: transient failures whose recovery
115
+ * window is either long (a resource still leaving a Pending/Creating state) or
116
+ * genuinely load-related (throttling), where hammering AWS with dense retries
117
+ * is harmful and exponential backoff is the correct shape.
118
+ *
119
+ * The name-cooldown spellings used to live here too (`wait 60 seconds`, S3's
120
+ * `conflicting conditional operation`); they moved to
121
+ * {@link NAME_COOLDOWN_ERROR_MESSAGE_PATTERNS} so the ordinary-create path and
122
+ * the delete-then-re-create sites read ONE list instead of two that drifted
123
+ * apart (issue [#2116](https://github.com/go-to-k/cdkd/issues/2116)).
118
124
  */
119
125
  const OTHER_TRANSIENT_ERROR_MESSAGE_PATTERNS = [
120
126
  "currently in the following state: Pending",
@@ -123,10 +129,8 @@ const OTHER_TRANSIENT_ERROR_MESSAGE_PATTERNS = [
123
129
  "DependencyViolation",
124
130
  "does not exist",
125
131
  "Schema is currently being altered",
126
- "conflicting conditional operation",
127
132
  "scheduled for deletion",
128
133
  "Could not deliver test message",
129
- "wait 60 seconds",
130
134
  "concurrent update operation",
131
135
  "because it is in use",
132
136
  "Rate exceeded",
@@ -135,15 +139,101 @@ const OTHER_TRANSIENT_ERROR_MESSAGE_PATTERNS = [
135
139
  "Unable to complete operation due to concurrent modification"
136
140
  ];
137
141
  /**
142
+ * The **name-cooldown** third of {@link RETRYABLE_ERROR_MESSAGE_PATTERNS}:
143
+ * an AWS service that holds a resource's NAME (or other unique identifier)
144
+ * while an ASYNCHRONOUS delete of the previous holder is still in flight, so a
145
+ * create of the same name inside that window is refused with a
146
+ * service-specific message. The window always clears on its own — the delete
147
+ * that opened it is already running — which is what makes every entry here
148
+ * retryable rather than terminal, whichever call site hits it.
149
+ *
150
+ * Read by BOTH consumers, which is the whole point of the list existing
151
+ * (issue [#2116](https://github.com/go-to-k/cdkd/issues/2116)):
152
+ *
153
+ * - {@link isNameCooldownError}, and through it
154
+ * {@link isRecreateRetryableError}, the retry filter at the
155
+ * delete-then-re-create sites (the deploy engine's `--replace` delete-first
156
+ * fallback, the recreate-via-* path, the rollback executor's
157
+ * delete-new-first) — the sites where cdkd itself just deleted the name
158
+ * holder;
159
+ * - {@link RETRYABLE_ERROR_MESSAGE_PATTERNS}, i.e. the ORDINARY create path,
160
+ * where a fresh `cdkd deploy` process has no idea a prior `cdkd destroy`
161
+ * deleted anything.
162
+ *
163
+ * The second consumer is the reachable one and was the measured failure in
164
+ * #2116: destroy-then-redeploy is a routine dev loop, CloudFormation absorbs
165
+ * the window and converges, and cdkd claims template compatibility with
166
+ * CloudFormation — so failing the whole deploy (26 resources created and
167
+ * rolled back, on the run that filed the issue) over a condition that clears
168
+ * in seconds is a parity defect. Before this list, the two consumers held
169
+ * DIFFERENT spellings of the same SQS error: the wire message
170
+ * (`wait 60 seconds`) was in the generic table so an ordinary create retried
171
+ * it, while the error CODE (`QueueDeletedRecently`) was not — so whether the
172
+ * identical AWS condition was survivable depended on which spelling the SDK
173
+ * happened to surface.
174
+ *
175
+ * Bounded, not unbounded: since #2116 the ordinary-create path rides
176
+ * `withRetry`'s NAME-COOLDOWN grid (8 retries, 2s/4s/8s then capped at 10s ≈
177
+ * 64s of sleep — `NAME_COOLDOWN_INITIAL_DELAY_MS` in `./retry.ts`), which is
178
+ * the same budget the re-create sites already carried. It is deliberately NOT
179
+ * the generic 47s schedule this path used to inherit: SQS's own sentence names
180
+ * a 60-second window, so 47s would not converge, it would merely fail 47s
181
+ * later. A name that is NOT in fact being
182
+ * released — a genuine, permanent collision — is a different signature
183
+ * ({@link isNameCollisionError}) and is deliberately NOT here, so it still
184
+ * fails fast into the actionable `--replace` refusal instead of burning a
185
+ * budget it cannot survive.
186
+ *
187
+ * **What must NOT go in this list.** Every entry below is specific to a delete
188
+ * that is ALREADY in flight and clears within a budget. Three candidates the
189
+ * sibling sweep turned up are recorded here as deliberate EXCLUSIONS rather
190
+ * than left unmentioned, because "absent" and "considered and rejected" are
191
+ * indistinguishable to the next person doing this sweep:
192
+ *
193
+ * - **ELBv2 `DuplicateLoadBalancerName`** and **DynamoDB's create-side
194
+ * `Table already exists: <name>`** — AWS raises both for a resource that
195
+ * merely EXISTS, just as readily as for a deleting one, so neither is
196
+ * distinguishable from a terminal collision. Promoting either would convert
197
+ * a fast, actionable `--replace` refusal into a full retry budget ending in
198
+ * the same failure. (DynamoDB's `Table is being deleted` IS distinguishable,
199
+ * but promoting THAT one re-multiplies the destroy-runner budget arithmetic
200
+ * `src/provisioning/dynamodb-index-busy-delete.ts` derives against the
201
+ * per-resource deadline — a different reason, so it is stated separately
202
+ * rather than folded in with the two above.)
203
+ * - **Secrets Manager `scheduled for deletion`** — the same one-sided shape as
204
+ * S3's entry (generic table only, invisible to
205
+ * {@link isRecreateRetryableError}), and the provider does delete with
206
+ * `ForceDeleteWithoutRecovery: true`, so it LOOKS like it belongs. It is
207
+ * excluded because that single message covers TWO conditions with wildly
208
+ * different windows: a force-deleted secret's name releasing in
209
+ * seconds-to-minutes, and a secret scheduled for deletion with a
210
+ * `RecoveryWindowInDays` of 7-30 DAYS, which no bounded budget can ride out
211
+ * and which a user reaches by deleting a secret outside cdkd. A budget that
212
+ * cannot converge on half its population is worse than failing fast, so the
213
+ * entry stays where it already was — in the generic table, retryable on an
214
+ * ordinary create and terminal at the re-create sites, unchanged by #2116.
215
+ */
216
+ const NAME_COOLDOWN_ERROR_MESSAGE_PATTERNS = [
217
+ "QueueDeletedRecently",
218
+ "wait 60 seconds",
219
+ "StateMachineDeleting",
220
+ "State Machine is being deleted",
221
+ "conflicting conditional operation"
222
+ ];
223
+ /**
138
224
  * Patterns that mark an AWS error as a transient/retryable failure.
139
225
  * Each entry is a substring match against the error message; all of these
140
226
  * are situations where the same call typically succeeds after a short delay
141
227
  * because of eventual consistency or just-created-dependency propagation.
142
228
  *
143
- * Composed from the two halves above so retryability has ONE source of truth
229
+ * Composed from the three halves above so retryability has ONE source of truth
144
230
  * while `withRetry` can still pick a per-class backoff cadence.
145
231
  */
146
- const RETRYABLE_ERROR_MESSAGE_PATTERNS = [...IAM_PROPAGATION_ERROR_MESSAGE_PATTERNS, ...OTHER_TRANSIENT_ERROR_MESSAGE_PATTERNS];
232
+ const RETRYABLE_ERROR_MESSAGE_PATTERNS = [
233
+ ...IAM_PROPAGATION_ERROR_MESSAGE_PATTERNS,
234
+ ...OTHER_TRANSIENT_ERROR_MESSAGE_PATTERNS,
235
+ ...NAME_COOLDOWN_ERROR_MESSAGE_PATTERNS
236
+ ];
147
237
  /**
148
238
  * HTTP status codes that always indicate a transient failure worth retrying.
149
239
  * 429 = Too Many Requests (throttle), 503 = Service Unavailable.
@@ -575,26 +665,31 @@ function isNameCollisionError(message) {
575
665
  return /(?<!\b(?:must|not|should|may|cannot)\s)already exists?\b/i.test(message) || message.includes("AlreadyExists");
576
666
  }
577
667
  /**
578
- * Match the SQS same-name re-creation cooldown: after `DeleteQueue`, creating
579
- * a queue with the SAME name inside ~60s fails with
580
- * `AWS.SimpleQueueService.QueueDeletedRecently` ("You must wait 60 seconds
581
- * after deleting a queue before you can create another with the same name").
582
- *
583
- * The generic transient table above already carries 'wait 60 seconds' for
584
- * plain CREATEs (rapid destroy → redeploy loops), but the delete-then-re-create
585
- * sites (the deploy engine's --replace delete-first fallback and the rollback
586
- * executor's reverse-replacement) override the retry filter with
587
- * {@link isNameCollisionError}, which this signature does NOT match — so a
588
- * replacement revert used to fail fast mid-flight with the resource absent
589
- * from both AWS and state (issue #1206). Those sites now OR this matcher into
590
- * their retry filter, with a schedule long enough to cover the 60s window.
668
+ * Match a same-name re-creation cooldown an AWS service still holding a
669
+ * resource's name while its asynchronous delete finishes. Every recognised
670
+ * spelling lives in {@link NAME_COOLDOWN_ERROR_MESSAGE_PATTERNS}; add new ones
671
+ * THERE rather than here, so the ordinary-create path
672
+ * ({@link RETRYABLE_ERROR_MESSAGE_PATTERNS}, which composes that list) cannot
673
+ * drift out of step with the delete-then-re-create sites again. That drift is
674
+ * exactly what issue [#2116](https://github.com/go-to-k/cdkd/issues/2116)
675
+ * found: SQS's wire message was retryable on an ordinary create while its
676
+ * error code was not, and the Step Functions spelling was recognised by
677
+ * neither.
678
+ *
679
+ * The delete-then-re-create sites (the deploy engine's `--replace`
680
+ * delete-first fallback and the rollback executor's reverse-replacement)
681
+ * override the retry filter with {@link isNameCollisionError}, which these
682
+ * signatures do NOT match — so a replacement revert used to fail fast
683
+ * mid-flight with the resource absent from both AWS and state (issue #1206).
684
+ * Those sites OR this matcher into their retry filter, with a schedule long
685
+ * enough to cover the 60s SQS window.
591
686
  *
592
687
  * Kept separate from {@link isNameCollisionError} on purpose: a cooldown at a
593
688
  * create-first site must NOT be treated as a collision (deleting the new
594
689
  * resource would not release the cooldown on the old name).
595
690
  */
596
691
  function isNameCooldownError(message) {
597
- return message.includes("QueueDeletedRecently") || message.includes("wait 60 seconds");
692
+ return NAME_COOLDOWN_ERROR_MESSAGE_PATTERNS.some((p) => message.includes(p));
598
693
  }
599
694
  /**
600
695
  * Retry filter for the delete-then-re-create sites: the old name holder was
@@ -989,8 +1084,10 @@ var PartialFailureError = class PartialFailureError extends CdkdError {
989
1084
  * is the whole point: the message interpolates the resource's LOGICAL ID,
990
1085
  * and the retry classifiers match by SUBSTRING, so a perfectly ordinary
991
1086
  * composite CDK id like `MyDependencyViolationSub` made
992
- * `isRetryableTransientError` return true (`DependencyViolation` is the only
993
- * whitespace-free entry in `RETRYABLE_ERROR_MESSAGE_PATTERNS`) and burned the
1087
+ * `isRetryableTransientError` return true (`DependencyViolation` was then the
1088
+ * only whitespace-free entry in `RETRYABLE_ERROR_MESSAGE_PATTERNS`; the
1089
+ * name-cooldown error CODES `QueueDeletedRecently` / `StateMachineDeleting`
1090
+ * joined it under issue #2116, so the hazard is wider, not narrower) and burned the
994
1091
  * full generic schedule — 8 retries, ~47s of pure sleep — on a path that was
995
1092
  * never going to succeed, before the fallback the error exists to trigger was
996
1093
  * even reached.
@@ -7575,6 +7672,45 @@ var S3StateBackend = class {
7575
7672
  return keys;
7576
7673
  }
7577
7674
  /**
7675
+ * Raw sidecar-object listing WITH metadata under the state bucket.
7676
+ *
7677
+ * {@link listRawKeys}'s twin, and it exists because an age-guarded sweep
7678
+ * needs `LastModified` and the reclaim plan needs `Size` — neither of which a
7679
+ * key list carries. Used by `cdkd gc`'s custom-resource-response sweep
7680
+ * (issue #2052), where the age is the only thing separating an abandoned
7681
+ * placeholder from one a concurrent run is about to write to.
7682
+ *
7683
+ * `LastModified` / `Size` are omitted from the response only for a key S3
7684
+ * did not return metadata for, which does not happen for `ListObjectsV2`
7685
+ * `Contents` entries; an entry missing either is DROPPED rather than
7686
+ * defaulted, because defaulting the date would either exempt an object from
7687
+ * the age guard forever or expose it immediately, and both are wrong in a
7688
+ * direction the caller cannot see.
7689
+ */
7690
+ async listRawObjects(keyPrefix) {
7691
+ await this.ensureClientForBucket();
7692
+ const objects = [];
7693
+ let continuationToken;
7694
+ do {
7695
+ const response = await this.s3Client.send(new ListObjectsV2Command({
7696
+ Bucket: this.config.bucket,
7697
+ ...await this.ownerParam(),
7698
+ Prefix: keyPrefix,
7699
+ ...continuationToken && { ContinuationToken: continuationToken }
7700
+ }));
7701
+ for (const obj of response.Contents ?? []) {
7702
+ if (obj.Key === void 0 || obj.LastModified === void 0 || obj.Size === void 0) continue;
7703
+ objects.push({
7704
+ key: obj.Key,
7705
+ lastModified: obj.LastModified,
7706
+ size: obj.Size
7707
+ });
7708
+ }
7709
+ continuationToken = response.IsTruncated ? response.NextContinuationToken : void 0;
7710
+ } while (continuationToken);
7711
+ return objects;
7712
+ }
7713
+ /**
7578
7714
  * Raw sidecar-object batch delete under the state bucket. Used by the
7579
7715
  * deployment-events pruner (issue #885) to drop superseded `{runId}.jsonl`
7580
7716
  * streams + their index. Chunked to the 1,000-key `DeleteObjects` ceiling.
@@ -9760,6 +9896,68 @@ const IAM_PROPAGATION_MAX_DELAY_MS = 2e3;
9760
9896
  * generic one by more than 0.75s in the early band.
9761
9897
  */
9762
9898
  const IAM_PROPAGATION_MAX_RETRIES = 26;
9899
+ /**
9900
+ * Backoff for the NAME-COOLDOWN class on the DEFAULT schedule — an AWS service
9901
+ * still holding a resource's name while an asynchronous delete of the previous
9902
+ * holder finishes (`NAME_COOLDOWN_ERROR_MESSAGE_PATTERNS` in
9903
+ * ./retryable-errors.ts). Issue
9904
+ * [#2116](https://github.com/go-to-k/cdkd/issues/2116).
9905
+ *
9906
+ * **Why this class does not inherit the generic schedule.** The generic one
9907
+ * sleeps 1+2+4+8+8+8+8+8 = 47s, and the longest window in this class NAMES its
9908
+ * own duration: SQS's own sentence is "You must wait 60 seconds after deleting
9909
+ * a queue before you can create another with the same name." A 47s budget
9910
+ * against a 60s window does not converge — it just fails 47s later, which is
9911
+ * strictly worse than failing at once. That mismatch was live for as long as
9912
+ * `wait 60 seconds` has been in the generic table.
9913
+ *
9914
+ * **Where the numbers come from — precedent, not a fresh guess.** They are the
9915
+ * delete-then-re-create sites' existing budget, chosen for this exact window
9916
+ * (`deploy-engine.ts`'s `--replace` fallback and `rollback-executor.ts`'s
9917
+ * reverse-replacement both pass `maxRetries: 8, initialDelayMs: 2_000,
9918
+ * maxDelayMs: 10_000`). Adopting it here makes the ordinary create path and
9919
+ * the re-create sites ride the SAME window with the SAME budget, which is the
9920
+ * property #2116 is about: whether a cooldown is survivable must not depend on
9921
+ * which call site met it.
9922
+ *
9923
+ * 2 + 4 + 8 + 10 x 5 = 64s over 8 retries
9924
+ *
9925
+ * Probe grid (seconds after the first failure): 2, 6, 14, 24, 34, 44, 54, 64 —
9926
+ * covering SQS's 60s and, with room to spare, the ~23s of `status: DELETING`
9927
+ * measured on an idle Step Functions state machine
9928
+ * (`tests/integration/custom-resource-provider`).
9929
+ *
9930
+ * **What it does NOT claim.** A window LONGER than 64s still fails, with the
9931
+ * same AWS message as before. This class converts "always fails" into "fails
9932
+ * only past a budget derived from the longest window AWS itself documents" —
9933
+ * and, unlike the generic schedule, a run that exhausts it now says so (see
9934
+ * the give-up summary in {@link withRetry}), so a too-short budget is
9935
+ * reportable rather than indistinguishable from having no retry at all.
9936
+ *
9937
+ * The retry COUNT is deliberately the generic 8 rather than a new ceiling, so
9938
+ * nothing about the loop's own exit condition moves. Same reason the dense IAM
9939
+ * grid needed its own count and this does not.
9940
+ *
9941
+ * **But "only the delay grid changes" understates the effect at the NESTED
9942
+ * sites, so state the compounding rather than implying there is none.** Three
9943
+ * call sites wrap a DEFAULT-schedule `withRetry` inside their own outer loop
9944
+ * (`deploy-engine.ts`'s two `--replace` / recreate sites and
9945
+ * `rollback-executor.ts`'s reverse-replacement). The inner loop is the one
9946
+ * this grid changes, and the outer one re-enters it per attempt, so the
9947
+ * product grows with it: total sleep on a cooldown at those sites measures
9948
+ * **487s -> 640s (8.1 -> 10.7 min)**. That is accepted, not overlooked -- it
9949
+ * stays inside the 30-minute per-resource `withResourceDeadline`, and the
9950
+ * alternative is an inner loop that cannot ride out the window it is nested
9951
+ * for. A reader auditing those budgets needs the number, which is why it is
9952
+ * here and not left to be re-derived.
9953
+ *
9954
+ * `dynamodb-index-busy-delete.ts` is unaffected and that was checked rather
9955
+ * than assumed: it passes all four knobs, so `defaultSchedule` is false and
9956
+ * this grid never applies to it -- its own deadline arithmetic stands.
9957
+ */
9958
+ const NAME_COOLDOWN_INITIAL_DELAY_MS = 2e3;
9959
+ /** Cap for the name-cooldown backoff. See {@link NAME_COOLDOWN_INITIAL_DELAY_MS}. */
9960
+ const NAME_COOLDOWN_MAX_DELAY_MS = 1e4;
9763
9961
  const defaultSleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
9764
9962
  /**
9765
9963
  * Run `operation`, retrying transient failures with exponential backoff
@@ -9779,6 +9977,13 @@ const defaultSleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
9779
9977
  * re-create sites' ~64s budget covering SQS's 60s name cooldown) and gets it
9780
9978
  * verbatim.
9781
9979
  *
9980
+ * A THIRD class rides its own grid on the default schedule since issue
9981
+ * [#2116](https://github.com/go-to-k/cdkd/issues/2116): a NAME COOLDOWN (a
9982
+ * service still holding a name while an async delete finishes) gets
9983
+ * 2s/4s/8s then 10s to a 64s total, because the longest window in that class
9984
+ * NAMES its duration -- SQS says "wait 60 seconds" -- and the generic 47s
9985
+ * cannot ride out a 60s wait. See {@link NAME_COOLDOWN_INITIAL_DELAY_MS}.
9986
+ *
9782
9987
  * The class is re-evaluated per attempt, so a propagation retry that runs into
9783
9988
  * a throttle backs OFF exponentially for that attempt instead of hammering.
9784
9989
  *
@@ -9809,6 +10014,8 @@ async function withRetry(operation, logicalId, opts = {}) {
9809
10014
  let sawPropagation = false;
9810
10015
  let propagationRetries = 0;
9811
10016
  let propagationSleptMs = 0;
10017
+ let nameCooldownRetries = 0;
10018
+ let nameCooldownSleptMs = 0;
9812
10019
  let serverErrorRetries = 0;
9813
10020
  for (let attempt = 0; attempt <= attemptCeiling; attempt++) try {
9814
10021
  return await operation();
@@ -9819,12 +10026,15 @@ async function withRetry(operation, logicalId, opts = {}) {
9819
10026
  const retryable = opts.isRetryable ? opts.isRetryable(message, error) : isRetryableTransientError(error, message);
9820
10027
  const propagation = defaultSchedule && isIamPropagationError(message);
9821
10028
  if (propagation) sawPropagation = true;
10029
+ const nameCooldown = defaultSchedule && !propagation && isNameCooldownError(message);
9822
10030
  const attemptLimit = sawPropagation ? 26 : maxRetries;
9823
10031
  if (!retryable || attempt >= attemptLimit) {
9824
- if (propagationRetries > 0 || serverErrorRetries > 0) {
10032
+ if (propagationRetries > 0 || serverErrorRetries > 0 || nameCooldownRetries > 0) {
9825
10033
  const budgetExhausted = sawPropagation && attempt >= attemptLimit;
10034
+ const nameCooldownBudgetExhausted = nameCooldownRetries >= maxRetries && attempt >= attemptLimit;
9826
10035
  const spent = [];
9827
10036
  if (propagationRetries > 0) spent.push(`${propagationRetries} IAM-propagation ${propagationRetries === 1 ? "retry" : "retries"} over ${(propagationSleptMs / 1e3).toFixed(2)}s of propagation backoff${budgetExhausted ? " (the full propagation budget)" : ""}`);
10037
+ if (nameCooldownRetries > 0) spent.push(`${nameCooldownRetries} name-cooldown ${nameCooldownRetries === 1 ? "retry" : "retries"} over ${(nameCooldownSleptMs / 1e3).toFixed(2)}s waiting for the name to be released${nameCooldownBudgetExhausted ? " (the full name-cooldown budget)" : ""}`);
9828
10038
  if (serverErrorRetries > 0) spent.push(`${serverErrorRetries} transient server-error ${serverErrorRetries === 1 ? "retry" : "retries"} (HTTP 5xx)`);
9829
10039
  const summary = () => `${logicalId}: gave up after ${spent.join(" and ")} - ${message}` + formatRetryClassificationSignals(error);
9830
10040
  try {
@@ -9833,7 +10043,7 @@ async function withRetry(operation, logicalId, opts = {}) {
9833
10043
  }
9834
10044
  throw error;
9835
10045
  }
9836
- const delay = propagation ? Math.min(250 * Math.pow(2, attempt), IAM_PROPAGATION_MAX_DELAY_MS) : Math.min(initialDelayMs * Math.pow(2, attempt), maxDelayMs);
10046
+ const delay = propagation ? Math.min(250 * Math.pow(2, attempt), IAM_PROPAGATION_MAX_DELAY_MS) : nameCooldown ? Math.min(NAME_COOLDOWN_INITIAL_DELAY_MS * Math.pow(2, attempt), NAME_COOLDOWN_MAX_DELAY_MS) : Math.min(initialDelayMs * Math.pow(2, attempt), maxDelayMs);
9837
10047
  const backoffThroughThisAttemptMs = propagation ? propagationSleptMs + delay : propagationSleptMs;
9838
10048
  opts.logger?.debug(` ⏳ Retrying ${logicalId} in ${delay / 1e3}s (attempt ${attempt + 1}/${attemptLimit}${propagation ? `, ${(backoffThroughThisAttemptMs / 1e3).toFixed(2)}s backoff through this attempt` : ""}) - ${message}`);
9839
10049
  for (let waited = 0; waited < delay; waited += 1e3) {
@@ -9843,6 +10053,9 @@ async function withRetry(operation, logicalId, opts = {}) {
9843
10053
  if (propagation) {
9844
10054
  propagationRetries++;
9845
10055
  propagationSleptMs = backoffThroughThisAttemptMs;
10056
+ } else if (nameCooldown) {
10057
+ nameCooldownRetries++;
10058
+ nameCooldownSleptMs += delay;
9846
10059
  } else if (opts.isRetryable === void 0 && isTransientServerError(error)) serverErrorRetries++;
9847
10060
  }
9848
10061
  throw lastError;
@@ -16261,7 +16474,8 @@ var IntrinsicFunctionResolver = class IntrinsicFunctionResolver {
16261
16474
  * criterion `retryable-errors.ts` documents as insufficient: the classifiers
16262
16475
  * match by SUBSTRING, and `sourceClause` interpolates template-controlled
16263
16476
  * text, so a logical id like `MyDependencyViolationHandler` puts
16264
- * `DependencyViolation` (the table's only whitespace-free entry) into the
16477
+ * `DependencyViolation` (a whitespace-free entry in the table — the only
16478
+ * one until issue #2116 added the name-cooldown error codes) into the
16265
16479
  * message. Reachability is real even though resolution runs outside
16266
16480
  * `withRetry` on the flat path: `NestedStackProvider.create` runs a child
16267
16481
  * `DeployEngine.deploy()` and re-throws, and the parent wraps `create()` in
@@ -18276,7 +18490,7 @@ var CloudControlProvider = class {
18276
18490
  if (context?.finalSnapshotIdentifier !== void 0) throw new ProvisioningError(`${logicalId} (${resourceType}) requires a final snapshot (DeletionPolicy: Snapshot), but the Cloud Control API delete route has no final-snapshot parameter. Re-run with --skip-final-snapshot after snapshotting manually, or retain the resource.`, resourceType, logicalId, physicalId);
18277
18491
  if (context?.removeProtection === true && resourceType === "AWS::AutoScaling::AutoScalingGroup") {
18278
18492
  this.logger.debug(`Delegating protected AutoScalingGroup ${logicalId} delete to the SDK ASGProvider (Cloud Control cannot force-delete a protected ASG)`);
18279
- const { ASGProvider } = await import("./asg-provider-DYnIz-Gy.js").then((n) => n.n);
18493
+ const { ASGProvider } = await import("./asg-provider-BaNTIjU4.js").then((n) => n.n);
18280
18494
  return await new ASGProvider().delete(logicalId, physicalId, resourceType, _properties, context);
18281
18495
  }
18282
18496
  const isProtectedEc2Instance = context?.removeProtection === true && resourceType === "AWS::EC2::Instance";
@@ -19221,6 +19435,42 @@ function endCommandInterruptScope() {
19221
19435
  resetInterruptWatchLatch();
19222
19436
  }
19223
19437
 
19438
+ //#endregion
19439
+ //#region src/state/state-prefix.ts
19440
+ /**
19441
+ * The default S3 key prefix for cdkd state.
19442
+ *
19443
+ * Homed in the STATE layer rather than in `src/cli/commands/state-file-keys.ts`
19444
+ * (which re-exports it, so its four existing importers are unchanged) because
19445
+ * `src/state/lock-contention-message.ts` needs it to decide whether a recovery
19446
+ * hint should spell `--state-prefix` at all, and a `src/state/**` module
19447
+ * importing from `src/cli/commands/**` inverts the layering — the CLI sits
19448
+ * ABOVE the state layer in the 7-layer architecture, not below it.
19449
+ *
19450
+ * Note this is only the DEFAULT. Other commands accept `--state-prefix`, so
19451
+ * whole-bucket listings deliberately do not scope to it.
19452
+ */
19453
+ const DEFAULT_STATE_PREFIX = "cdkd";
19454
+ /**
19455
+ * The state-bucket prefix `CustomResourceProvider` PUTs its response
19456
+ * placeholders under, one object per invocation
19457
+ * (`custom-resource-responses/{requestId}.json`).
19458
+ *
19459
+ * Homed here for the same layering reason as {@link DEFAULT_STATE_PREFIX}: the
19460
+ * PRODUCER is `src/provisioning/providers/custom-resource-provider.ts` and the
19461
+ * COLLECTOR is `src/cli/commands/gc.ts`, so a copy in either would be a copy
19462
+ * the other could drift from — and the two spellings would then disagree about
19463
+ * which objects exist, which is the only way a sweeper can miss the family it
19464
+ * was written for (issue #2052). `src/cli/commands/state-file-keys.ts`
19465
+ * re-exports it so gc reads it alongside the other state-key constants.
19466
+ *
19467
+ * Note this is only the DEFAULT: `ProviderRegistry` can be configured with a
19468
+ * different `responsePrefix`, so a sweep scoped to this value is a sweep of the
19469
+ * default layout. gc has no access to a non-default one — nothing persists it —
19470
+ * which is stated at the sweep's own call site rather than implied here.
19471
+ */
19472
+ const CUSTOM_RESOURCE_RESPONSE_PREFIX = "custom-resource-responses";
19473
+
19224
19474
  //#endregion
19225
19475
  //#region src/provisioning/providers/custom-resource-provider.ts
19226
19476
  /**
@@ -26075,7 +26325,8 @@ async function replayRollback(operations, stateResources, stackName, ctx, option
26075
26325
  * hang the flag exists to prevent. Those providers retry internally.
26076
26326
  * - **Interrupt.** `replayRollback` polls interrupts only BETWEEN ops, so an
26077
26327
  * un-threaded `isInterrupted` leaves Ctrl-C dead for the length of the
26078
- * backoff schedule (~47s) per op.
26328
+ * backoff schedule per op — ~47s on the generic grid, or ~64s if the op
26329
+ * hits a name cooldown, which rides its own longer grid since issue #2116.
26079
26330
  *
26080
26331
  * Returns the provider's result so the caller can honour
26081
26332
  * `effectiveProperties` (issue #1644) — both revert arms used to write the
@@ -26381,8 +26632,13 @@ async function updateWithRollbackRetry(provider, args, logicalId, logger, isInte
26381
26632
  * on attempt 0 and it reaches the caller's catch on the FIRST outer attempt,
26382
26633
  * exactly as before. The SQS cooldown IS matched by the inner classifier (the
26383
26634
  * generic table carries `wait 60 seconds`), which is the same division of
26384
- * labour the deploy engine's named-replacement site documents: the inner retry
26385
- * absorbs most of the 60s window and the outer ~64s budget covers the tail.
26635
+ * labour the deploy engine's named-replacement site documents. Since issue
26636
+ * #2116 the inner retry rides the name-cooldown grid (≈64s) rather than the
26637
+ * generic ~47s one, so it covers the whole 60s window on its own instead of
26638
+ * absorbing most of it and leaving a tail for the outer loop; the outer loop
26639
+ * now earns its place by ALSO covering the late name release that the inner
26640
+ * default classifier rejects. The two compound — measured at 640s of total
26641
+ * sleep on a cooldown, inside the 30-minute per-resource deadline.
26386
26642
  */
26387
26643
  async function createWithRollbackRetry(provider, create, logicalId, logger, isInterrupted, secrets, outer) {
26388
26644
  if (provider.disableOuterRetry) return await create();
@@ -29615,5 +29871,5 @@ var DeployEngine = class {
29615
29871
  };
29616
29872
 
29617
29873
  //#endregion
29618
- export { startInterruptWatch as $, AwsClients as $n, shouldRetainResource as $t, renderStatefulReason as A, getLegacyStateBucketName as An, isThrottlingError as Ar, redactSecretsForState as At, exportAliasCollisionScrubWarning as B, CFN_TEMPLATE_BODY_LIMIT as Bn, DiffCalculator as Bt, isFinalSnapshotError as C, runDockerForeground as Cn, SynthesisError as Cr, TEMPLATE_SOURCED_RULES as Ct, extractDeploymentEventError as D, Synthesizer as Dn, withErrorHandling as Dr, isSingleDynamicReferenceToken as Dt, makeCanonicalizePropertiesFn as E, getDockerImageBySourceHash as En, normalizeAwsError as Er, errorCauseChain as Et, green as F, resolveStateBucketWithDefault as Fn, s3BucketDomainName as Ft, collectInlinePolicyNamesManagedBySiblings as G, expectedOwnerParam as Gn, TemplateParser as Gt, secretBearingStateKeyWarning as H, MIGRATE_TMP_PREFIX as Hn, describeTypeWithThrottleRetry as Ht, red as I, resolveStateBucketWithDefaultAndSource as In, s3BucketDualStackDomainName as It, findActionableSilentDrops as J, derivePartitionAndUrlSuffix as Jn, S3StateBackend as Jt, clearOnUpdateRemoval as K, PARTITION_TABLE as Kn, LockManager as Kt, yellow as L, resolveUseCdkBootstrapAssets as Ln, s3BucketRegionalDomainName as Lt, bold as M, resolveAutoAssetStorage as Mn, __exportAll as Mr, classifyReplaySecretRegion as Mt, cyan as N, resolveCaptureObservedState as Nn, producerRegionsFromState as Nt, MULTI_REGION_RECREATE_BLOCKED_TYPES as O, synthesisStatusMessage as On, isMarkedNonRetryable as Or, maskSecretsInError as Ot, gray as P, resolveSkipPrefix as Pn, s3BucketArn as Pt, isInterruptedWaitError as Q, resolveBucketRegion as Qn, importableOutputs as Qt, collectDeclaredOutputNames as R, stateBucketExistenceConfirmed as Rn, s3BucketWebsiteUrl as Rt, createPreDeleteFinalSnapshot as S, partitionSensitiveEnv as Sn, StateError as Sr, STATE_SOURCED_READBACK_RULES as St, unsupportedFinalSnapshotError as T, AssetManifestLoader as Tn, isCdkdError as Tr, dynamicReferenceTokens as Tt, stateKeySecretExposure as U, findLargeInlineResources as Un, withRetry as Ut, isExportAliasCollision as V, CFN_TEMPLATE_URL_LIMIT as Vn, INTRINSIC_KEYS as Vt, IAMRoleProvider as W, uploadCfnTemplate as Wn, DagBuilder as Wt, beginCommandInterruptScope as X, processStackMessages as Xn, exportNamesCarriedFrom as Xt, findSilentDropProperties as Y, AssemblyReader as Yn, rebuildClientForBucketRegion as Yt, endCommandInterruptScope as Z, clearBucketRegionCache as Zn, importableOutputKeys as Zt, computeImplicitDeleteEdges as _, buildDenyExternalAccessPolicy as _n, ProvisioningError as _r, replayWarn as _t, DeploymentEventsStore as a, loadPublishableAssetManifest as an, ConfigError as ar, carriesDynamicReference as at, buildFinalSnapshotIdentifier as b, formatDockerLoginError as bn, StackHasActiveImportsError as br, requireConfigString as bt, replayFailedOperations as c, stripControlChars as cn, DeployCancelledError as cr, refStateLookupFromResource as ct, updatePartialReason as d, ensureAssetStorage as dn, LocalMigrateError as dr, resolveExplicitPhysicalId as dt, AssetPublisher as en, getAwsClients as er, CloudControlProvider as et, UNSPECIFIED_SKIP_REASON as f, getBootstrapMarkerKey as fn, LocalStartServiceError as fr, assertRegionMatch as ft, IMPLICIT_DELETE_DEPENDENCIES as g, validateContainerRepoName as gn, PartialFailureError as gr, readConfigString as gt, maskingRetryLogger as h, validateAssetBucketName as hn, NestedStackChildDirectDestroyError as hr, configStringRefusal as ht, DeploymentEventsReader as i, createAssetRedirectResolver as in, CdkdError as ir, IntrinsicFunctionResolver as it, formatResourceLine as j, resolveApp as jn, markNonRetryable as jr, scrubResourceRecord as jt, isStatefulRecreateTargetSync as k, getDefaultStateBucketName as kn, isRetryableTransientError as kr, maskSecretsInText as kt, replayRollback as l, AssetModeResolver as ln, DynamicReferenceRegionAmbiguousError as lr, WAFv2WebACLProvider as lt, withResourceDeadline as m, readBootstrapMarkerBody as mn, MissingCdkCliError as mr, configBooleanRefusal as mt, DEFAULT_RESOURCE_WARN_AFTER_MS as n, WorkGraph as nn, setAwsClients as nr, disableInstanceApiTermination as nt, planFailedOps as o, rewriteTemplateAssetReferences as on, CrossAccountSecretRefusalError as or, cfnRefValueFromPhysicalId as ot, deleteSkipReason as p, parseBootstrapMarker as pn, LockError as pr, coerceCfnBoolean as pt, ProviderRegistry as q, canonicalizeRegion as qn, displaySafe as qt, DeployEngine as r, buildAssetRedirectMap as rn, AssetError as rr, isTerminationProtectionPropagationError as rt, planRollback as s, escapeRegExp$1 as sn, DependencyError as sr, getAccountInfo as st, DEFAULT_RESOURCE_TIMEOUT_MS as t, stringifyValue as tn, resetAwsClients as tr, slowCcOperationTimeoutMs as tt, updatePartialMessage as u, BOOTSTRAP_MARKER_PREFIX as un, LocalInvokeBuildError as ur, normalizeAwsTagsToCfn as ut, ATOMIC_FINAL_SNAPSHOT_TYPES as v, buildDockerImage as vn, ResourceTimeoutError as vr, requireConfigArray as vt, refusesFinalSnapshot as w, runDockerStreaming as wn, formatError as wr, createSecretMasker as wt, ccRoutedFinalSnapshotError as x, getDockerCmd as xn, StackTerminationProtectionError as xr, STATE_SOURCED_CROSS_GENERATION_RULES as xt, PRE_DELETE_SNAPSHOT_TYPES as y, dockerSpawnEnvWithSensitive as yn, ResourceUpdateNotSupportedError as yr, requireConfigObject as yt, collectPublishedOutputNames as z, warnDeprecatedNoPrefixCliFlag as zn, applyRoleArnIfSet as zt };
29619
- //# sourceMappingURL=deploy-engine-DxboGDr4.js.map
29874
+ export { endCommandInterruptScope as $, clearBucketRegionCache as $n, importableOutputKeys as $t, renderStatefulReason as A, synthesisStatusMessage as An, isMarkedNonRetryable as Ar, maskSecretsInError as At, exportAliasCollisionScrubWarning as B, stateBucketExistenceConfirmed as Bn, s3BucketWebsiteUrl as Bt, isFinalSnapshotError as C, getDockerCmd as Cn, StackTerminationProtectionError as Cr, STATE_SOURCED_CROSS_GENERATION_RULES as Ct, extractDeploymentEventError as D, AssetManifestLoader as Dn, isCdkdError as Dr, dynamicReferenceTokens as Dt, makeCanonicalizePropertiesFn as E, runDockerStreaming as En, formatError as Er, createSecretMasker as Et, green as F, resolveCaptureObservedState as Fn, producerRegionsFromState as Ft, collectInlinePolicyNamesManagedBySiblings as G, findLargeInlineResources as Gn, withRetry as Gt, secretBearingStateKeyWarning as H, CFN_TEMPLATE_BODY_LIMIT as Hn, DiffCalculator as Ht, red as I, resolveSkipPrefix as In, s3BucketArn as It, findActionableSilentDrops as J, PARTITION_TABLE as Jn, LockManager as Jt, clearOnUpdateRemoval as K, uploadCfnTemplate as Kn, DagBuilder as Kt, yellow as L, resolveStateBucketWithDefault as Ln, s3BucketDomainName as Lt, bold as M, getLegacyStateBucketName as Mn, isThrottlingError as Mr, redactSecretsForState as Mt, cyan as N, resolveApp as Nn, markNonRetryable as Nr, scrubResourceRecord as Nt, MULTI_REGION_RECREATE_BLOCKED_TYPES as O, getDockerImageBySourceHash as On, normalizeAwsError as Or, errorCauseChain as Ot, gray as P, resolveAutoAssetStorage as Pn, __exportAll as Pr, classifyReplaySecretRegion as Pt, beginCommandInterruptScope as Q, processStackMessages as Qn, exportNamesCarriedFrom as Qt, collectDeclaredOutputNames as R, resolveStateBucketWithDefaultAndSource as Rn, s3BucketDualStackDomainName as Rt, createPreDeleteFinalSnapshot as S, formatDockerLoginError as Sn, StackHasActiveImportsError as Sr, requireConfigString as St, unsupportedFinalSnapshotError as T, runDockerForeground as Tn, SynthesisError as Tr, TEMPLATE_SOURCED_RULES as Tt, stateKeySecretExposure as U, CFN_TEMPLATE_URL_LIMIT as Un, INTRINSIC_KEYS as Ut, isExportAliasCollision as V, warnDeprecatedNoPrefixCliFlag as Vn, applyRoleArnIfSet as Vt, IAMRoleProvider as W, MIGRATE_TMP_PREFIX as Wn, describeTypeWithThrottleRetry as Wt, CUSTOM_RESOURCE_RESPONSE_PREFIX as X, derivePartitionAndUrlSuffix as Xn, S3StateBackend as Xt, findSilentDropProperties as Y, canonicalizeRegion as Yn, displaySafe as Yt, DEFAULT_STATE_PREFIX as Z, AssemblyReader as Zn, rebuildClientForBucketRegion as Zt, computeImplicitDeleteEdges as _, validateAssetBucketName as _n, NestedStackChildDirectDestroyError as _r, configStringRefusal as _t, DeploymentEventsStore as a, buildAssetRedirectMap as an, AssetError as ar, isTerminationProtectionPropagationError as at, buildFinalSnapshotIdentifier as b, buildDockerImage as bn, ResourceTimeoutError as br, requireConfigArray as bt, replayFailedOperations as c, rewriteTemplateAssetReferences as cn, CrossAccountSecretRefusalError as cr, cfnRefValueFromPhysicalId as ct, updatePartialReason as d, AssetModeResolver as dn, DynamicReferenceRegionAmbiguousError as dr, WAFv2WebACLProvider as dt, importableOutputs as en, resolveBucketRegion as er, isInterruptedWaitError as et, UNSPECIFIED_SKIP_REASON as f, BOOTSTRAP_MARKER_PREFIX as fn, LocalInvokeBuildError as fr, normalizeAwsTagsToCfn as ft, IMPLICIT_DELETE_DEPENDENCIES as g, readBootstrapMarkerBody as gn, MissingCdkCliError as gr, configBooleanRefusal as gt, maskingRetryLogger as h, parseBootstrapMarker as hn, LockError as hr, coerceCfnBoolean as ht, DeploymentEventsReader as i, WorkGraph as in, setAwsClients as ir, disableInstanceApiTermination as it, formatResourceLine as j, getDefaultStateBucketName as jn, isRetryableTransientError as jr, maskSecretsInText as jt, isStatefulRecreateTargetSync as k, Synthesizer as kn, withErrorHandling as kr, isSingleDynamicReferenceToken as kt, replayRollback as l, escapeRegExp$1 as ln, DependencyError as lr, getAccountInfo as lt, withResourceDeadline as m, getBootstrapMarkerKey as mn, LocalStartServiceError as mr, assertRegionMatch as mt, DEFAULT_RESOURCE_WARN_AFTER_MS as n, AssetPublisher as nn, getAwsClients as nr, CloudControlProvider as nt, planFailedOps as o, createAssetRedirectResolver as on, CdkdError as or, IntrinsicFunctionResolver as ot, deleteSkipReason as p, ensureAssetStorage as pn, LocalMigrateError as pr, resolveExplicitPhysicalId as pt, ProviderRegistry as q, expectedOwnerParam as qn, TemplateParser as qt, DeployEngine as r, stringifyValue as rn, resetAwsClients as rr, slowCcOperationTimeoutMs as rt, planRollback as s, loadPublishableAssetManifest as sn, ConfigError as sr, carriesDynamicReference as st, DEFAULT_RESOURCE_TIMEOUT_MS as t, shouldRetainResource as tn, AwsClients as tr, startInterruptWatch as tt, updatePartialMessage as u, stripControlChars as un, DeployCancelledError as ur, refStateLookupFromResource as ut, ATOMIC_FINAL_SNAPSHOT_TYPES as v, validateContainerRepoName as vn, PartialFailureError as vr, readConfigString as vt, refusesFinalSnapshot as w, partitionSensitiveEnv as wn, StateError as wr, STATE_SOURCED_READBACK_RULES as wt, ccRoutedFinalSnapshotError as x, dockerSpawnEnvWithSensitive as xn, ResourceUpdateNotSupportedError as xr, requireConfigObject as xt, PRE_DELETE_SNAPSHOT_TYPES as y, buildDenyExternalAccessPolicy as yn, ProvisioningError as yr, replayWarn as yt, collectPublishedOutputNames as z, resolveUseCdkBootstrapAssets as zn, s3BucketRegionalDomainName as zt };
29875
+ //# sourceMappingURL=deploy-engine-B9SrF2-R.js.map