Auto defragment append blobs
Append blobs are great, however, when a such a blob is generated through a long series of small increments, the read performance of the resulting blob is very poor: from 10x to 20x slower than reading a regular page blob.
The performance problem goes away if the app rewrite the append-blob in large 4MB chunks. However, this process is complicated to setup, and collides with any 'always-on' property of the app.
As append blobs are append-only, it would be much better if Azure was taking care of defragmenting the append blobs on its own; possible through a dedicated API operation that would trigger the operation.
Thank you for your feedback. This work is planned for the coming year. For any further questions, or to discuss your specific scenario, send us an email at firstname.lastname@example.org.
Rob Eisenberg commented
Any update on this?
Victor Nicollet commented
We benchmarked a full download of two different append blobs. The first was created with a single append of 4MB, the second was created with 16000 appends of 250 bytes each (which corresponds to our typical usage), which is also about 4MB.
A full download of the first blob, from our fiber connection, takes 750 milliseconds on average. The second blob takes 21 seconds on average, or a 28x increase.