Fix #542: add a Citrus integration test for couchbase-source

[oscerd]

Follow-up to #3007, which shipped couchbase-source without a test on the grounds that the project's Citrus toolchain had no Couchbase container. That was true only in the sense that there is no built-in one — the generic container works fine, it just needs setting up by hand.

The test

testcontainers: start: container (couchbase/server:community-7.6.2)
  -> initCluster.groovy   (cluster init, bucket, primary index, seed one document)
  -> camel: cli: run      (couchbase-source -> log-sink)
  -> camel: cli: verify   (logMessage: "hello-from-couchbase")

Same shape as the existing mongodb-source-route test, so it should read familiarly.

Three things that were not optional

Recording these because each cost a full run to find, and the next person adding a container-based test will hit at least one.

Ports are bound one-to-one. The Couchbase SDK bootstraps from the management port and then reconnects using the addresses the cluster advertises. With a remapped host port the client bootstraps and then cannot reach the data service. Hence explicit 8091:8091, 8093:8093, 11210:11210 rather than exposedPorts.

The wait strategy needs a full-line regex. Testcontainers matches log wait patterns against the entire line, so "Starting Couchbase Server" never matches the actual line and the container times out after five minutes. It has to be ".*Starting Couchbase Server.*".

No Groovy GStrings in the script. Citrus resolves dollar-brace expressions in these scripts as its own test variables, so a GString fails with Unknown variable 'user'. The script uses string concatenation throughout — including, eventually, in the comment that explains why.

Why the statement looks the way it does

couchbase.statement=SELECT META().id AS __id, * FROM _default

Not the obvious SELECT * FROM <bucket>, because both halves matter:

• CouchbaseConsumer reads row.getString("__id") and skips any row without it, logging a warning rather than failing. A query without META().id AS __id yields a source that polls happily and emits nothing.
• The consumer runs the query through scope.query(...), so the keyspace is resolved as <bucket>._default.<what-you-wrote>. FROM <bucket> becomes <bucket>._default.<bucket> and fails with Keyspace not found.

I found both by running the Kamelet against a live cluster, and they mean the example currently shipped in couchbase-source is wrong — SELECT * FROM \travel-sample` LIMIT 10` would silently return nothing. That is my error from #3007. I am fixing it in a separate PR rather than folding a Kamelet change into a test PR.

Verification

Passes locally, twice through (test and verify phases), about 35 seconds each:

Tests run: 1, Failures: 0, Errors: 0, Skipped: 0 -- in CouchbaseIT
  ✔ SUCCESS (38088ms) couchbase-source-route-test

mvn clean install passes with tests from the repository root.

For reviewers

The image is large — couchbase/server:community-7.6.2 is about 1.7 GB, materially bigger than redpanda or floci. On a cold CI cache that is a real cost for one test, and worth weighing against the value of covering this Kamelet.

I did not add camel.apache.org/kamelet.verified: "true". The contributor guide ties that label to having passing behaviour tests, but it is applied to only 21 Kamelets and notably not to mongodb-source, which does have a test — so it looks like a curation decision rather than an automatic consequence. Happy to add it if you want it applied here.

---

Claude Code on behalf of Andrea Cosentino

[oscerd]

CI is green, and the new test genuinely ran rather than being skipped:

Tests run: 1, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 61.15 s -- in CouchbaseIT

Adding the cost figure to go with the image-size caveat above, since it is the number that matters for the decision: 61 seconds on the CI runner, against 35 locally — the difference being the image pull on a cold cache.

For context, that puts it mid-pack rather than at the top:

CommonIT     398.5 s
AwsIT        408.3 s
AvroIT        86.99 s
KafkaIT       80.15 s
ProtobufIT    79.20 s
JiraIT        69.43 s
CouchbaseIT   61.15 s
MongoIT       50.78 s

So the 1.7 GB image costs about a minute of a suite that already runs for well over twenty. Still your call whether one Kamelet is worth that, but it is a smaller marginal cost than the image size alone suggests.

---

Claude Code on behalf of Andrea Cosentino