IoT hub with kafka connector - how to decode the enqueued timestamp and device id

[Pálmi]

I'm reading data from the default endpoint of an IoT hub in azure using the kafka connector in Databricks.  Most data items are straight forward, but the device id and the timestamp I haven't been able to properly decode

For example, the key-value map of the headers  {"key": "iothub-enqueuedtime", "value": "gwAAAZAMsGjg"}  should be a recent timestamp.  Any ideas on how to decode this, using pyspark?

[Kaniz_Fatma]

Hi @Pálmi, 

1. First, read the data stream from your IoT hub using the Kafka connector. You can define a DataFrame to read the data as a stream from your EventHub or IoT-Hub. 
2. To extract the timestamp, you’ll need to define the structure of the data.
3. The timestamp provided in the iothub-enqueuedtime header is in a base64-encoded format. You can decode it and convert it to a readable timestamp. 
4. Finally, select the relevant columns (including the decoded timestamp) from the DataFrame.

For more information, you can refer to the the Microsoft Learn guide on Azure IoT Hub message format.

[Pálmi]

Hi @Kaniz_Fatma , thanks for your reply.  The iothub-enqueuedtime does not (directly ) cast into a timestamp, but an unix timestamp with milliseconds  is somewhere in there

from pyspark.sql import SparkSession

from pyspark.sql.functions import col, from_json, explode, get_json_object, schema_of_json

from pyspark.sql.functions import col, explode, expr, unbase64, from_unixtime,hex,length

df = spark.read.format("delta").table("iot_ps2")

#df.display()

df2=df.select("headers")

df2.display()

# Explode the array of structs into individual rows

df_exploded = df.withColumn("json_item", explode(col("headers")))

# Filter rows to get only the 'iothub-enqueuedtime' key

df_filtered = df_exploded.filter(col("json_item.key") == "iothub-enqueuedtime")

df3=df_filtered.select("json_item.key","json_item.value")

df3=df3.withColumn("str_value",expr("cast(value as STRING)"))

df3=df3.withColumn("hex",expr("hex(str_value)"))

df3.display()

looking at the hex code it is possible to determine that the 6 rightmost bytes "01 90 36 4C 1B 5C" turn into a unix timestamp with milliseconds. That leaves 3 unknown bytes 

I'm hoping that a more straightforward way is available

[Erik]

https://github.com/Azure/azure-event-hubs-for-kafka/issues/56#issuecomment-1432006831