Claus Ibsen created CAMEL-26235:
-----------------------------------
Summary: camel-hazelcast - with no operation header and no
defaultOperation the producers fail with a NullPointerException, and the map
consumer never gets TTL expiry events
Key: CAMEL-26235
URL: https://issues.apache.org/jira/browse/CAMEL-26235
Project: Camel
Issue Type: Bug
Components: camel-hazelcast
Reporter: Claus Ibsen
Fix For: 4.24.0
Found while adding the catalog body types of the eleven hazelcast components
(CAMEL-24961, PR #28396).
h4. 1. No operation: NullPointerException instead of a clear error (map,
multimap, replicatedmap, list, set, ringbuffer, pncounter)
{{HazelcastComponentHelper.lookupOperation}} (main 2a465aa4814c, line 63)
returns the endpoint's {{defaultOperation}}, which is null when not set, and
the producers switch on it ({{HazelcastMapProducer}} line 76,
{{HazelcastMultimapProducer}} line 53, {{HazelcastReplicatedmapProducer}} line
54, {{HazelcastListProducer}} line 61, {{HazelcastSetProducer}} line 48,
{{HazelcastRingbufferProducer}} line 43, {{HazelcastPNCounterProducer}} line
43), so a message without {{CamelHazelcastOperationType}} fails with a
NullPointerException ('"operation" is null'). Checked on 4.24.0-SNAPSHOT with
an embedded Hazelcast 5.4 cluster.
Proposed: fail with a message that names the {{CamelHazelcastOperationType}}
header and the {{defaultOperation}} option.
h4. 2. map consumer: TTL expiry gives no event
{{CamelMapListener}} (line 27) implements {{MapEntryListener}}, which extends
the added, updated, removed, evicted, merged and cleared listeners but not
{{EntryExpiredListener}} ({{MapEntryListener}} lines 27-30), so an entry that
expires by TTL never reaches the route. The multimap and replicatedmap
{{CamelEntryListener}} handles {{entryExpired}}. Read in the code, not run.
Proposed: add {{EntryExpiredListener}} to {{MapEntryListener}} and send an
expired event as the other listener does.
h4. 3. list: get without CamelHazelcastObjectIndex fails with a
NullPointerException
{{HazelcastListProducer.get}} (line 121) calls {{list.get(pos)}} with the
{{Integer}} from the {{CamelHazelcastObjectIndex}} header (line 56); without
the header {{pos}} is null and unboxing throws. Checked on 4.24.0-SNAPSHOT with
an embedded Hazelcast 5.4 cluster.
Proposed: fail with a message that the get operation needs the
{{CamelHazelcastObjectIndex}} header.
h4. 4. list and set: getAll puts the live proxy in the body
{{HazelcastListProducer.getAll}} (line 169) and {{HazelcastSetProducer.getAll}}
(line 118) set the distributed {{IList}} / {{ISet}} proxy as the body, not a
copy, so later steps see (and can change) the live collection. Checked on
4.24.0-SNAPSHOT with an embedded Hazelcast 5.4 cluster.
Proposed: set a copy ({{new ArrayList<>(list)}}, {{new HashSet<>(set)}}).
h4. 5. queue: removeIf always fails, drainTo without its header passes null
{{HazelcastQueueProducer.removeIf}} (lines 165-167) calls
{{queue.removeIf(filter)}}, which the Hazelcast queue proxy does not support
(UnsupportedOperationException "remove() is not supported!"), so the operation
can never work (seen). {{drainTo}} (lines 96-97, 155) passes the
{{CamelHazelcastDrainToCollection}} header as is, so without it
{{queue.drainTo(null)}} fails (read in the code). Checked on 4.24.0-SNAPSHOT
with an embedded Hazelcast 5.4 cluster.
Proposed: implement removeIf by iterating and removing the matching items (or
remove the operation), and fail with a clear message when the drainTo
collection header is missing.
h4. 6. pncounter: getAndAdd with no body fails with a NullPointerException
{{HazelcastPNCounterProducer.getAndAdd}} (line 88) unboxes
{{getBody(Long.class)}} into a {{long}}. Checked on 4.24.0-SNAPSHOT with an
embedded Hazelcast 5.4 cluster.
Proposed: fail with a message that getAndAdd needs a numeric body.
h4. 7. seda: a body that is neither Serializable nor convertible to bytes gives
an unclear error
{{HazelcastSedaProducer.checkAndStore}} (lines 60-67) converts a
non-Serializable body with {{getBody(byte[].class)}}; when that gives null,
{{queue.add(null)}} fails with "Null item is not allowed!". Checked on
4.24.0-SNAPSHOT with an embedded Hazelcast 5.4 cluster.
Proposed: fail with a message that the body must be Serializable or convertible
to bytes.
h4. 8. Docs: replicatedmap intro and old header name in the samples
{{hazelcast-replicatedmap-component.adoc}} starts with the intro of the
instance component ("consume join/leave events of the cache instance in the
cluster"), and the XML and YAML samples of the replicatedmap, multimap,
pncounter and ringbuffer pages set the old header {{hazelcast.operation.type}}
instead of {{CamelHazelcastOperationType}}. Not fixed in the PR.
Proposed: fix the intro and use {{CamelHazelcastOperationType}} in the samples.
_Claude Code on behalf of davsclaus_
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
