Local Functions

Note

This page is for developers of Azure Functions - not users

Introduction¶

Running Azure Functions locally is a little involved; this guide has been written to get you up and running with Azure CLI and Functions. Note that this guide assumes you're using VSCode. All of the functionality described here can be replicated outside of VSCode (in Azure CLI) but instructions for doing so are not detailed in this guide.

Note that this guide is not intended to demonstrate how to create an Azure Function from scratch. Its purpose is to show you how to run pre-coded Functions.

To follow this guide you need to install Azure CLI.

Azure Functions Basics¶

Conceptually, queue-based Azure Functions are simple to follow. A message is posted to a message queue. A Function then executes on messages within a message queue. In Carrot-Mapper, a message is posted to a message queue as JSON but other formats are possible (e.g. XML). Here's an example message destined for the scanreports queue:

{
  "scan_report_id": 105, 
  "blob_name": "GOSH_CoStar_WhiteRabbit_MetaData_V24_T29kygl.xlsx"
}

And here is an example message destined for the nlpqueue queue:

{
  "documents": 
  [{
    "language": "en", 
    "id": "17730_field", 
    "text": "The patient has a fever"
  }]
}

An Azure function is made up of three key files. They are kept in the function's directory (app/workers/...) and define what and how the function should run:

init.py - This contains the function's logic. The method main() should be present for the function to execute.
function.json - This tells Azure what type of function you're developing and also what queue from which to pick messages up from.

Another file is kept outside of the function's directory and should not be commited to Git:

local.settings.json - Contains environment variables for your function. It must be present to run Azure functions locally. Contact a member of the development team to get all keys and connection strings for this file. For reference, a local.settings.json file looks like this:

{
  "IsEncrypted": false,
  "Values": {
    "AzureWebJobsStorage": "REDACTED,
    "FUNCTIONS_WORKER_RUNTIME": "python",
    "STORAGE_CONN_STRING": "REDACTED",
    "NLP_API_KEY": "REDACTED",
    "APP_URL": "REDACTED",
    "AZ_FUNCTION_KEY": "REDACTED",
    "SCAN_REPORT_QUEUE_NAME": "scanreports-local",
    "NLP_QUEUE_NAME": "nlpqueue-local"

  }
}

You only need to have 1 local.settings.json file for all functions within the Carrot-Mapper project. Note, however, that it should be updated with extra queue names if/when the project requires them.

Azure Functions in Carrot-Mapper¶

The Carrot-Mapper project has several different Azure Functions: one for processing scan reports (called ProcessQueue which posts to the message queue called scanreports), another to run the NLP service (called NLPQueue which posts to the message queue called nlpqueue), and one to generate mapping rules for a scan report. The code for these functions lives in the /workers directory.

Azure Functions Queues¶

For each process (NLP and Scan Reports), we have two queues each - a 'local' queue and a 'live' queue (for a total of 4 unique queues.) As environment variables, they are encoded as:

Local queues¶

NLP_QUEUE_NAME=nlpqueue-local
SCAN_REPORT_QUEUE_NAME=scanreports-local

Deployed queues¶

NLP_QUEUE_NAME=nlpqueue
SCAN_REPORT_QUEUE_NAME=scanreports

The purpose of the local queues is that, when developing locally, messages are sent only to a 'local' queue.

These environment variables which control where messages are sent must be maintained in the following locations:

local.settings.json - A local file to hold Azure Function environment variables. Should point to 'local' (e.g. NLP_QUEUE_NAME=nlpqueue-local).
Within the Carrot webapp, the .env file must include variables for 'local' on your local machine and set to 'live' variables on App Service.

Running Azure Functions Locally¶

We use Azurite to enable running the functions locally, emulating the Azure Function services.

To start debugging locally in VSCode you must first ensure that Carrot is up and running locally (see here for building and running the Carrot-Mapper Docker image ). Once your local server is running, you can start Azure Function's debugging mode by clicking:

Run (top toolbar) -> Start Debugging

After a few seconds, you'll see something like the following printed to the debugging terminal that's started when you run debugging:

> Executing task: . .venv/bin/activate && func host start <

Found Python version 3.9.5 (python3).

Azure Functions Core Tools
Core Tools Version:       3.0.3477 Commit hash: 5fbb9a76fc00e4168f2cc90d6ff0afe5373afc6d  (64-bit)
Function Runtime Version: 3.0.15584.0

[2021-07-02T09:34:28.929Z] Cannot create directory for shared memory usage: /dev/shm/AzureFunctions
[2021-07-02T09:34:28.929Z] System.IO.FileSystem: Access to the path '/dev/shm/AzureFunctions' is denied. Operation not permitted.

Functions:

        NLPQueue: queueTrigger
        ProcessQueue: queueTrigger

For detailed output, run func with --verbose flag.
[2021-07-02T09:34:32.711Z] Traceback (most recent call last):
[2021-07-02T09:34:32.711Z]   File "/usr/local/Cellar/azure-functions-core-tools@3/3.0.3477/workers/python/3.9/OSX/X64/azure_functions_worker/bindings/shared_memory_data_transfer/file_accessor_unix.py", line 127, in _get_valid_mem_map_dirs
[2021-07-02T09:34:32.711Z]     os.makedirs(dir_path)
[2021-07-02T09:34:32.711Z]   File "/usr/local/Cellar/python@3.9/3.9.5/Frameworks/Python.framework/Versions/3.9/lib/python3.9/os.py", line 215, in makedirs
[2021-07-02T09:34:32.711Z]     makedirs(head, exist_ok=exist_ok)
[2021-07-02T09:34:32.711Z]   File "/usr/local/Cellar/python@3.9/3.9.5/Frameworks/Python.framework/Versions/3.9/lib/python3.9/os.py", line 225, in makedirs
[2021-07-02T09:34:32.711Z]     mkdir(name, mode)
[2021-07-02T09:34:32.711Z] PermissionError: [Errno 1] Operation not permitted: '/dev/shm'
[2021-07-02T09:34:32.812Z] Worker process started and initialized.
[2021-07-02T09:34:36.731Z] Host lock lease acquired by instance ID '000000000000000000000000DC4198B8'.

You'll likely see a few errors/warnings about directories and permissions. At the time of writing this guide, it's safe to ignore these! Note that two functions are listed as you start debugging mode: NLPQueue and ProcessQueue. These are the names of the directories which hold the function's code, not the name of the message queue which holds the function's messages. The text after the function name (queueTrigger) specifies the type of function it is (defined in function.json). For more information on Function types, click here.

With the final console output saying 'Host lock lease acquired', you're ready to run your first Azure Functions job!

Running an NLP task¶

The remainder of this guide will show what to expect to see in the debugging console if you run NLP at the field level.

Navigate to a field within a scan report.
Ensure the field you've chosen has a meaningful description (e.g. "Patient has a cold"). If it doesn't have one, click 'Edit Field' and manually put one in.
Ensure that pass_from_source is set to True within the field's settings. This will cause CCOM to process the field only.
Click 'Run NLP'
After a short wait a green message banner will appear to say NLP is now running.

Hop back to VSCode. After a few seconds, you should see the debugging console updating to say that it has received the message from the message queue:

Executing 'Functions.NLPQueue' (Reason='New queue message detected on 'nlpqueue-local'.', Id=ea27d6be-eb0b-490c-9cab-82cf2a119355)
Trigger Details: MessageId: f00a438f-9bad-4110-bb8b-5278461d7bb3, DequeueCount: 1, InsertionTime: 2021-07-02T10:57:26.000+00:00

If you have configured .env and local.settings.json correctly, you will see that the job was posted to the queue nlpqueue-local.

After some time (around 15-30 seconds), you should notice further updates to the debugging console. Due to various print() statements in the init.py file (remember, this holds the function's logic), the console will start logging what's occurring as the function processes the message in the message queue. First, the console shows you what message was sent to the queue:

MESSAGE >>>  {'id': 'f00a438f-9bad-4110-bb8b-5278461d7bb3', 'body': '{"documents": [{"language": "en", "id": "17569_field", "text": "patient has a cold"}]}'}

After job submission, the NLPQueue function works through the stages detailed in NLP Processing. Having received concept codes from the NLP API, the function looks up a standard and valid conceptID using the CCOM API. Sometimes, a concept code will be returned which doesn't have a standard and valid conceptID. In such instances you will see the following in the debugging console:

Concept Code C34500 not found!

However, if a standard and valid conceptID is found, the Azure function then saves the data to the ScanReportConcept model. This is shown to you in the debugging console with the PAYLOAD print statement like so:

SAVING TO FIELD LEVEL...
PAYLOAD >>> {'nlp_entity': 'cold', 'nlp_entity_type': 'SymptomOrSign', 'nlp_confidence': 0.75, 'nlp_vocabulary': 'ICD10CM', 'nlp_concept_code': 'J00', 'concept': 260427, 'object_id': '17569', 'content_type': 15}
SAVING TO FIELD LEVEL...
PAYLOAD >>> {'nlp_entity': 'cold', 'nlp_entity_type': 'SymptomOrSign', 'nlp_confidence': 0.75, 'nlp_vocabulary': 'ICD9CM', 'nlp_concept_code': '460', 'concept': 260427, 'object_id': '17569', 'content_type': 15}
SAVING TO FIELD LEVEL...
PAYLOAD >>> {'nlp_entity': 'cold', 'nlp_entity_type': 'SymptomOrSign', 'nlp_confidence': 0.75, 'nlp_vocabulary': 'SNOMEDCT_US', 'nlp_concept_code': '82272006', 'concept': 260427, 'object_id': '17569', 'content_type': 15}

Here, three conceptIDs were found for three different vocabularies (ICD10CM, ICD9CM and SNOWMEDCT_US). Therefore, three records are saved to ScanReportConcepts (hence the three SAVING TO FIELD LEVEL... statements)

After successful execution, the debug console will finally tell you that the job has finished:

Executed 'Functions.NLPQueue' (Succeeded, Id=ea27d6be-eb0b-490c-9cab-82cf2a119355, Duration=30234ms)

Here is a full interrupted trace of the above process:

Executing 'Functions.NLPQueue' (Reason='New queue message detected on 'nlpqueue-local'.', Id=ea27d6be-eb0b-490c-9cab-82cf2a119355)
Trigger Details: MessageId: f00a438f-9bad-4110-bb8b-5278461d7bb3, DequeueCount: 1, InsertionTime: 2021-07-02T10:57:26.000+00:00
MESSAGE >>>  {'id': 'f00a438f-9bad-4110-bb8b-5278461d7bb3', 'body': '{"documents": [{"language": "en", "id": "17569_field", "text": "patient has a cold"}]}'}
Concept Code C34500 not found!
SAVING TO FIELD LEVEL...
PAYLOAD >>> {'nlp_entity': 'cold', 'nlp_entity_type': 'SymptomOrSign', 'nlp_confidence': 0.75, 'nlp_vocabulary': 'ICD10CM', 'nlp_concept_code': 'J00', 'concept': 260427, 'object_id': '17569', 'content_type': 15}
SAVING TO FIELD LEVEL...
PAYLOAD >>> {'nlp_entity': 'cold', 'nlp_entity_type': 'SymptomOrSign', 'nlp_confidence': 0.75, 'nlp_vocabulary': 'ICD9CM', 'nlp_concept_code': '460', 'concept': 260427, 'object_id': '17569', 'content_type': 15}
SAVING TO FIELD LEVEL...
PAYLOAD >>> {'nlp_entity': 'cold', 'nlp_entity_type': 'SymptomOrSign', 'nlp_confidence': 0.75, 'nlp_vocabulary': 'SNOMEDCT_US', 'nlp_concept_code': '82272006', 'concept': 260427, 'object_id': '17569', 'content_type': 15}
Executed 'Functions.NLPQueue' (Succeeded, Id=ea27d6be-eb0b-490c-9cab-82cf2a119355, Duration=30234ms)

Stopping Debugging¶

Debugging can be stopped by clicking:

Run (top menu) -> Stop Debugging