addIoTDatabaseIndex races the IoT Agent and kills the services script

[kzangeli]

addIoTDatabaseIndex drops every collection in iotagentjson and then re-creates devices and groups:

db.getCollectionNames().forEach(c=>db[c].drop());
db.createCollection("devices");
...

By the time it runs, the IoT Agent container is already up and connected to the same database — it is started by the same docker compose up. If the agent creates devices in the window between the drop loop and the createCollection call, the call fails:

MongoServerError: Collection iotagentjson.devices already exists.

On mongo 6.0, which is what MONGO_DB_VERSION pins, db.createCollection on an existing collection throws NamespaceExists; mongosh then exits 1; and services runs under set -e, so the whole invocation dies — after the stack is up, before the tutorial has run a single request.

Because it depends on who wins the race, it shows up as an intermittent mystery red rather than as a bug. It was caught in tutorials.Big-Data-Flink's own CI, where the identical commit then passed on re-run.

(Worth knowing for later: mongo 8 made createCollection return {ok:1} in this case, so raising MONGO_DB_VERSION would hide this rather than fix it.)

The fix

The two createCollection calls are redundant — createIndex creates the collection implicitly when it is missing. Dropping them closes the window with no other change.

Verified on mongo 6.0:

	current	amended
clean database	exit 0	exit 0
run twice in a row	exit 1	exit 0
collection created concurrently mid-script	exit 1	exit 0

and the resulting indexes are identical either way:

devices: {_id}, {_id.service,_id.id,_id.type}, {_id.type}, {_id.id}
groups:  {_id}, {_id.resource,_id.apikey,_id.service}, {_id.type}

Scope

Nine tutorials carry this function verbatim. This is one of them; the same one-hunk change is going to each, with FIWARE/tutorials.Big-Data-Flink#17 as the original.

🤖 Generated with Claude Code