Trigger is not called periodically because of closed transaction passed to setTriggerMetadata

[dima-guzman]

Expected Behavior

A trigger is installed in the DB. It is expected it is reliable called each time a transaction has been committed that creates new nodes

Actual Behavior

Time to time when a series of queries is executed, the trigger execution code and in particular setTriggerMetadata mehtod receives a closed transaction, exception is thown and the trigger is not called

How to Reproduce the Problem

Intensively insert new nodes to the DB
The following example shows simultaneous insertion of some test nodes and relations. It is written is javascript, but any language that can send queries within transactions would allow to reporduce this

for (let i = 0; i < 15; i++) {
	setInterval(async () => {
		await runInTransaction(async (transaction) => {
			await transaction.run(
				`
WITH timestamp() as t
CREATE (r:Refund {id: "RF-1" + t + $i, isDeleted: false})
CREATE (r2:Refund {id: "RF-2" + t + $i, isDeleted: false})
CREATE (r3:Refund {id: "RF-3" + t + $i, isDeleted: false})
CREATE (r4:Refund {id: "RF-4" + t + $i, isDeleted: false})
CREATE (d:Document {id: "RI-1" + t + $i, isDeleted: false})
CREATE (d)-[:HAS_DOCUMENT]->(r)
CREATE (d)-[:HAS_DOCUMENT]->(r2)
CREATE (d)-[:HAS_DOCUMENT]->(r3)
CREATE (d)-[:HAS_DOCUMENT]->(r4)
				`,
				{ i }
			);
			console.info(`Thread ${i} insertion`);
		});
	}, 500);
}

There is also a trigger in the DB, the issue reproduces for both after and afterAsync stages

CALL apoc.trigger.install("${dbName}", "changeDetection", "
                ...
		RETURN 0
	", {phase: "afterAsync"})

Steps (Mandatory)

1. Install a trigger to the DB
2. Run Cypher queries creating nodes in parallel threads
3. A crash happens with error message in the logs

2025-07-22 08:03:52.662+0000 ERROR id:14528408 - transaction id:411382 - 0 ms: (planning: 0, waiting: 0) - 312 B - 0 page hits, 0 page faults - embedded-session                core.prd -  - CALL tx.setMetaData($data) - {data: {apoc.trigger: true}} - runtime=pipelined - {}
org.neo4j.exceptions.CypherExecutionException: Failed to invoke procedure `tx.setMetaData`: Caused by: org.neo4j.graphdb.NotInTransactionException: The transaction has been closed.
        at org.neo4j.cypher.internal.planning.ExceptionTranslatingReadQueryContext.callDbmsProcedure(ExceptionTranslatingQueryContext.scala:357) ~[neo4j-cypher-5.26.8.jar:5.26.8]
        at org.neo4j.cypher.internal.runtime.DbmsCallMode$.callProcedure(ProcedureCallMode.scala:107) ~[neo4j-cypher-runtime-util-5.26.8.jar:5.26.8]
        at org.neo4j.codegen.OperatorTaskPipeline0_F9691E16_21569.compiledOperate(Unknown Source) ~[?:?]
        at org.neo4j.cypher.internal.runtime.pipelined.operators.CompiledTask.operateWithProfile(OperatorCodeGenBaseTemplates.scala:705) ~[neo4j-cypher-pipelined-runtime-5.26.8.jar:5.26.8]
        at org.neo4j.cypher.internal.runtime.pipelined.RegularPipelineTask.executeOperators(PipelineTask.scala:128) ~

Specifications (Mandatory)

Versions

• OS: Windows 10, Ubuntu 22.04, Ubuntu 24.04
• Neo4j: 5.26.8
• Neo4j-Apoc: 5.26.8

[dima-guzman]

The solution we found for ourselves is a patch to APOC code base that retries a failed trigger call (failed also because of APOC internal reasons).
After 12h of test the issue is no more reproduced

diff --git a/core/src/main/java/apoc/trigger/TriggerHandler.java b/core/src/main/java/apoc/trigger/TriggerHandler.java
index 31bb4c10..76fcd082 100644
--- a/core/src/main/java/apoc/trigger/TriggerHandler.java
+++ b/core/src/main/java/apoc/trigger/TriggerHandler.java
@@ -61,6 +61,7 @@ public class TriggerHandler extends LifecycleAdapter implements TransactionEvent
     private static final Map<String, Object> TRIGGER_META = Map.of("apoc.trigger", true);

     public static final String TRIGGER_REFRESH = "apoc.trigger.refresh";
+    public static final String TRIGGER_NRETRIES = "apoc.trigger.nretries";

     // Snapshot of installed triggers. The containing map is immutable.
     private final AtomicReference<Map<String, Map<String, Object>>> triggersSnapshot = new AtomicReference<>(Map.of());
@@ -249,11 +250,11 @@ public class TriggerHandler extends LifecycleAdapter implements TransactionEvent
         }

         if (hasPhase(Phase.after)) {
-            try (Transaction tx = db.beginTx()) {
-                setTriggerMetadata(tx);
-                executeTriggers(tx, txData, Phase.after);
-                tx.commit();
-            }
+            Util.retryInTx(log, db, (inner) -> {
+                setTriggerMetadata(inner);
+                executeTriggers(inner, txData, Phase.after);
+                return null;
+            }, 0, apocConfig().getInt(TRIGGER_NRETRIES, 3), _ignored -> {});
         }
         afterAsync(txData);
     }
@@ -266,11 +267,11 @@ public class TriggerHandler extends LifecycleAdapter implements TransactionEvent
     private void afterAsync(TransactionData txData) {
         if (hasPhase(Phase.afterAsync)) {
             TriggerMetadata triggerMetadata = TriggerMetadata.from(txData, true);
-            Util.inTxFuture(pools.getDefaultExecutorService(), db, (inner) -> {
+            Util.inTxFuture(log, pools.getDefaultExecutorService(), db, (inner) -> {
                 setTriggerMetadata(inner);
                 executeTriggers(inner, triggerMetadata.rebind(inner), Phase.afterAsync);
                 return null;
-            });
+            }, apocConfig().getInt(TRIGGER_NRETRIES, 3), _ignored -> {}, _ignored -> {});
         }
     }

[dima-guzman]

Can I create a PR for this? I can set default number of retries to 0 so by default trigger behavior will not change.

[loveleif]

@dima-guzman thanks for the report, we have prioritized this and the plan is to have a look this or next week. But you're welcome to contribute a PR if you wish.

[loveleif]

@dima-guzman I'm having problems reproducing this. I have tried running 100k queries concurrently on all 16 cores of my machine and the triggers have all been successful. Do you have any more details about your test setup that might help me out? Thanks.