CosmosDB Serverless Change Feed not triggering Python Azure Function

Hi ,

Thanks for reaching out to Microsoft Q&A.

This is a known behaviour pattern with Cosmos DB change feed + AzFunctions (Python, especially on Flex Consumption): the trigger is checkpointing correctly (lease moving forward) but not invoking after the first batch due to host scaling/listener lifecycle issues or function not staying warm. In serverless + low throughput scenarios, the listener can go idle and not re-poll aggressively, even though the continuation token advances. The most common causes are: missing lease_container_name explicitly (leading to default mismatch), not setting create_lease_container_if_not_exists=True, low RU/s causing delayed polling, or Flex Consumption cold-start behaviour where the change feed processor is not continuously active. Also, Python worker+Cosmos trigger has intermittent reliability issues compared to .NET.

What typically fixes it? explicitly configure lease container + prefix, ensure max_items_per_invocation and feed_poll_delay are tuned, increase RU temporarily to validate, and most importantly switch to Elastic Premium plan or .NET function if this is production-critical. For Python, adding logging.info inside a loop over documents also helps validate batch processing (sometimes empty batches get swallowed silently).

Please 'Upvote'(Thumbs-up) and 'Accept' as answer if the reply was helpful. This will be benefitting other community members who face the same issue.

Hi @Craig Webb

The most interesting detail in your troubleshooting is that the lease continuation token and timestamp continue to advance when new documents are inserted. This indicates that the Cosmos DB Change Feed Processor is successfully reading changes from the container and updating the lease state.

Because of that, the issue is less likely to be with Cosmos DB itself and more likely related to the Azure Functions runtime, Cosmos DB trigger extension, or the Python worker process.

A few details would help narrow this down:

• Azure Functions runtime version
• Extension bundle version (or Cosmos DB extension version)
• Contents of your host.json
• Confirmation that the Cosmos DB account is using the SQL (Core) API and Serverless mode
• Any host restart, recycle, or scale events visible in Application Insights

Since you're running on a Flex Consumption plan, I'd also recommend checking Application Insights for worker restarts or host lifecycle events occurring after the first successful trigger execution.

The continuation token advancing is the key diagnostic clue here—it confirms that Cosmos DB is generating change feed events and that the trigger infrastructure is consuming them. The next step is determining why those events are not consistently reaching the Python function after the initial invocation.

Could you share the runtime/extension versions and your host.json configuration?

hi Craig Webb & thanks for join me here at Q&A portal,

if lease tokens move but ur function logs only once, the trigger is probably checkpointing changes but not invoking the handler correctly. Configure the lease container instead of relying on defaults. Cosmos trigger https://learn.microsoft.com/en-us/azure/azure-functions/functions-bindings-cosmosdb-v2-trigger

Add lease_container_name="leases",

create_lease_container_if_not_exists=True

Then either delete existing lease documents or use a new lease container name for testing.

Check Application Insights requests table, not only traces. If invocations exist but logs do not, it is logging. If leases move but no invocations exist, it is trigger/runtime behavior. https://learn.microsoft.com/en-us/azure/azure-functions/functions-reference-python

Stale lease state or trigger config issue, not Cosmos Change Feed itself.

rgds,

Alex

&

If my answer was helpful pls mark it and additional thx if u follow me at Q&A portal

Hello Craig !

Thank you for posting on MS Learn Q&A.

The first thing I would check is whether another function app, deployment slot, local process or previous deployment is using the same monitored container with the same lease container.

For Cosmos DB triggers, the lease container tracks progress and if multiple functions are configured for the same container, each one should use a dedicated lease container or a different lease prefix otherwise only one of the functions is triggered.

https://learn.microsoft.com/en-us/azure/azure-functions/functions-bindings-cosmosdb-v2-trigger

So I would make the lease container explicit and unique:

import logging
import azure.functions as func
app = func.FunctionApp()
@app.function_name(name="ProcessAssetsHistory")
@app.cosmos_db_trigger(
    arg_name="documents",
    database_name="%COSMOS_DATABASE_NAME%",
    container_name="site-assets-history",
    connection="CosmosDBConnection",
    lease_container_name="site-assets-history-leases",
    create_lease_container_if_not_exists="true"
)
def process_assets_history(documents: func.DocumentList) -> None:
    logging.info(
        "Trigger invoked: container=%s docs=%d",
        "site-assets-history",
        len(documents)
    )

Also make sure the lease container has partition key /id, because partitioned lease containers are required to use /id.

https://learn.microsoft.com/en-us/azure/cosmos-db/change-feed-functions

do not reuse the same leases container for multiple Cosmos DB trigger functions unless each function has a separate lease prefix or a separate lease container. Since your continuation token is advancing, another listener may be consuming the feed and checkpointing it before this function logs anything.

To troubleshoot, you can enable the Cosmos DB trigger host logs in host.json.