Merge pull request #9043 from sfc-gh-anoyes/anoyes/idempotency-status-json

Add .cluster.idempotency_ids to status json

documentation/sphinx/source/mr-status-json-schemas.rst.inc

@@ -902,6 +902,13 @@
          "num_tenants" : 1, // on data cluster, local count; on management cluster, total metacluster count
          "num_tenant_groups" : 10,
          "tenant_group_capacity" : 20,
+      },
+      "idempotency_ids" : {
+         "size_bytes" : 0, // An estimate of the current number of bytes used in the database to store idempotency ids.
+         "expired_version" : 0, // The commit status of a transaction whose commit version could be <= expired_version can no longer be determined.
+         "expired_age" : 0, // The age in seconds of expired_version.
+         "oldest_id_version" : 0, // The version of the oldest idempotency id still stored in the database.
+         "oldest_id_age" : 0 // The age in seconds of the oldest_id_version.
       }
    },
    "client":{

fdbclient/IdempotencyId.actor.cpp

@@ -232,6 +232,43 @@ ACTOR static Future<Optional<Key>> getBoundary(Reference<ReadYourWritesTransacti
 	return result.front().key;
 }
 
+ACTOR Future<JsonBuilderObject> getIdmpKeyStatus(Database db) {
+	state Reference<ReadYourWritesTransaction> tr = makeReference<ReadYourWritesTransaction>(db);
+	state int64_t size;
+	state IdempotencyIdsExpiredVersion expired;
+	state KeyBackedObjectProperty<IdempotencyIdsExpiredVersion, _Unversioned> expiredKey(idempotencyIdsExpiredVersion,
+	                                                                                     Unversioned());
+	state int64_t oldestIdVersion = 0;
+	state int64_t oldestIdTime = 0;
+	loop {
+		try {
+			tr->setOption(FDBTransactionOptions::READ_SYSTEM_KEYS);
+			tr->setOption(FDBTransactionOptions::READ_LOCK_AWARE);
+
+			wait(store(size, tr->getEstimatedRangeSizeBytes(idempotencyIdKeys)) &&
+			     store(expired, expiredKey.getD(tr)) &&
+			     success(getBoundary(tr, idempotencyIdKeys, Oldest::True, &oldestIdVersion, &oldestIdTime)));
+			JsonBuilderObject result;
+			result["size_bytes"] = size;
+			if (expired.expired != 0) {
+				result["expired_version"] = expired.expired;
+			}
+			if (expired.expiredTime != 0) {
+				result["expired_age"] = int64_t(now()) - expired.expiredTime;
+			}
+			if (oldestIdVersion != 0) {
+				result["oldest_id_version"] = oldestIdVersion;
+			}
+			if (oldestIdTime != 0) {
+				result["oldest_id_age"] = int64_t(now()) - oldestIdTime;
+			}
+			return result;
+		} catch (Error& e) {
+			wait(tr->onError(e));
+		}
+	}
+}
+
 ACTOR Future<Void> cleanIdempotencyIds(Database db, double minAgeSeconds) {
 	state int64_t idmpKeySize;
 	state int64_t candidateDeleteSize;
@@ -315,8 +352,10 @@ ACTOR Future<Void> cleanIdempotencyIds(Database db, double minAgeSeconds) {
 			if (!finalRange.empty()) {
 				tr->addReadConflictRange(finalRange);
 				tr->clear(finalRange);
-				tr->set(idempotencyIdsExpiredVersion,
-				        ObjectWriter::toValue(IdempotencyIdsExpiredVersion{ candidateDeleteVersion }, Unversioned()));
+				tr->set(
+				    idempotencyIdsExpiredVersion,
+				    ObjectWriter::toValue(IdempotencyIdsExpiredVersion{ candidateDeleteVersion, candidateDeleteTime },
+				                          Unversioned()));
 				TraceEvent("IdempotencyIdsCleanerAttempt")
 				    .detail("Range", finalRange.toString())
 				    .detail("IdmpKeySizeEstimate", idmpKeySize)
@@ -332,4 +371,4 @@ ACTOR Future<Void> cleanIdempotencyIds(Database db, double minAgeSeconds) {
 		}
 	}
 	return Void();
-}
+}

fdbclient/Schemas.cpp

@@ -602,8 +602,6 @@ const KeyRef JSONSchemas::statusSchema = R"statusSchema(
             "description":"abc"
          }
       ],
-)statusSchema"
-                                         R"statusSchema(
       "recovery_state":{
          "seconds_since_last_recovered":1,
          "required_resolvers":1,
@@ -976,6 +974,13 @@ const KeyRef JSONSchemas::statusSchema = R"statusSchema(
          "num_tenants":0,
          "num_tenant_groups":10,
          "tenant_group_capacity":20
+      },
+      "idempotency_ids":{
+         "size_bytes": 0,
+         "expired_version": 0,
+         "expired_age": 0,
+         "oldest_id_version": 0,
+         "oldest_id_age": 0
       }
    },
    "client":{

fdbclient/include/fdbclient/IdempotencyId.actor.h

@@ -29,6 +29,7 @@
 #pragma once
 
 #include "fdbclient/FDBTypes.h"
+#include "fdbclient/JsonBuilder.h"
 #include "fdbclient/PImpl.h"
 #include "flow/Arena.h"
 #include "flow/IRandom.h"
@@ -46,10 +47,11 @@ struct IdempotencyIdsExpiredVersion {
 	// Any version at or below expired might have had its idempotency id expired. Any version greater than `expired`
 	// definitely has not had it's idempotency id expired.
 	Version expired = 0;
+	int64_t expiredTime = 0;
 
 	template <class Archive>
 	void serialize(Archive& ar) {
-		serializer(ar, expired);
+		serializer(ar, expired, expiredTime);
 	}
 };
 
@@ -187,6 +189,8 @@ KeyRangeRef makeIdempotencySingleKeyRange(Arena& arena, Version version, uint8_t
 
 void decodeIdempotencyKey(KeyRef key, Version& commitVersion, uint8_t& highOrderBatchIndex);
 
+ACTOR Future<JsonBuilderObject> getIdmpKeyStatus(Database db);
+
 // Delete zero or more idempotency ids older than minAgeSeconds
 //
 // Normally idempotency ids are deleted as part of the normal commit process, so this only needs to clean ids that

fdbserver/Status.actor.cpp

@@ -3124,6 +3124,8 @@ ACTOR Future<StatusReply> clusterGetStatus(
 		state JsonBuilderObject recoveryStateStatus = wait(
 		    recoveryStateStatusFetcher(cx, ccWorker, mWorker, workers.size(), &status_incomplete_reasons, &statusCode));
 
+		state JsonBuilderObject idmpKeyStatus = wait(getIdmpKeyStatus(cx));
+
 		// machine metrics
 		state WorkerEvents mMetrics = workerEventsVec[0].present() ? workerEventsVec[0].get().first : WorkerEvents();
 		// process metrics
@@ -3505,6 +3507,8 @@ ACTOR Future<StatusReply> clusterGetStatus(
 		if (!recoveryStateStatus.empty())
 			statusObj["recovery_state"] = recoveryStateStatus;
 
+		statusObj["idempotency_ids"] = idmpKeyStatus;
+
 		// cluster messages subsection;
 		JsonBuilderArray clientIssuesArr = getClientIssuesAsMessages(clientStatus);
 		if (clientIssuesArr.size() > 0) {