Replies: 1 comment 1 reply
-
To account for the increase in blob costs, we've switched away from using blobs for partition management to using Azure Tables for partition management. This significantly reduced costs for various workloads. More information is available here. ![]() PM refers to "Partition Manager". PM V2 corresponds to blob-based partition management and PM V3 refers to the newer table-based partition manager. The new partition manager is enabled by default in newer versions of the Durable Functions extension. It can also be enabled manually in host.json. There are no planned optimizations in place for queue reads. However, users can control this to a certain degree by changing the frequency at which they poll the queues, also in host.json. The other thing to note is that we'll be announcing the initial availability of a "consumption" plan for the Durable Task Scheduler (DTS), which may be attractive to some as an alternative to the Azure Storage backend, particularly if Azure Storage costs are increasing. The DTS Consumption plan will be billed based on "actions" that are taken by orchestrations, and there won't be any costs associated with idle apps that are polling for work but not finding any. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Azure Storage GPv1 accounts are being retired. New GPv1 accounts won't be available beginning early next year, and existing accounts will be migrated to GPv2 in October 2026.
Azure Storage GPv2 accounts employ a pricing model that penalizes transaction-heavy applications heavily, which can hit durable functions pretty hard. Operation price for both blob reads and queue reads are both increased by ~10x. For low-volume apps that see little compute/networking cost, this could easily translate into a significant price hike.
I wish to know if there are any optimizations currently in place, being implemented, or planned to mitigate the upcoming cost impact.
Beta Was this translation helpful? Give feedback.
All reactions