Claus Ibsen created CAMEL-26194:
-----------------------------------
Summary: camel-zookeeper - a Map, POJO, Integer or null body
overwrites the znode with null data, and InOnly write failures are only logged
Key: CAMEL-26194
URL: https://issues.apache.org/jira/browse/CAMEL-26194
Project: Camel
Issue Type: Bug
Components: camel-zookeeper
Reporter: Claus Ibsen
Fix For: 4.24.0
Found while adding the catalog body types of zookeeper (CAMEL-24961, PR #28414).
h4. 1. A body that does not convert to byte[] overwrites the znode with null
{{ZooKeeperUtils.getPayloadFromExchange}} (main 2a465aa4814c, line 90) uses the
non-mandatory {{ExchangeHelper.convertToType}}. For a Map, a POJO, an Integer
or a null body the payload is null and the producer silently sets the znode
data to null. Checked on 4.24.0-SNAPSHOT with ZooKeeper from test-infra
(docker).
Proposed: use a mandatory conversion for a non-null body (keep null for an
explicit empty write if that is wanted).
h4. 2. InOnly writes are fire-and-forget: failures are only logged
For an InOnly exchange the producer writes asynchronously
({{ZooKeeperProducer.process}} line 97 to 100) and the callbacks only log
errors ({{AsyncSetDataCallback}} line 171 to 190, for example a failed create
of a missing node), so the exchange succeeds even when the write failed.
Checked on 4.24.0-SNAPSHOT with ZooKeeper from test-infra.
Proposed: write synchronously by default (or wait for the callback) and fail
the exchange on error.
h4. 3. InOut DELETE with create=true creates a missing znode
{{ZooKeeperProducer.synchronouslyDelete}} (line 239 to 247) calls
{{createNode}} when the node does not exist, with the body as data, so a DELETE
of a missing node with {{create=true}} creates it. Read in the code, not run.
Proposed: a DELETE of a missing node should do nothing (or report it), never
create it.
h4. 4. Without repeat=true the consumer emits only the initial data
The consumer registers a {{DataChangedOperation}} ({{ZooKeeperConsumer}} line
235), but without {{repeat=true}} it does not go on after the first event (line
177), so later data changes are not delivered. Checked on 4.24.0-SNAPSHOT with
ZooKeeper from test-infra.
Proposed: document that repeat=true is needed to follow changes (or make the
watch follow changes by default).
_Claude Code on behalf of davsclaus_
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
