Does the FPolicy Send Buffer need increased due to EAGAIN in ONTAP 9.8 or higher?
Applies to
- ONTAP 9.8 or newer
- FPolicy
Answer
No, in most cases
Additional Information
- ONTAP 9.7 and older had a rewind on EAGAIN at a fixed 2 seconds
- The 2 second window created a loop:
- FPolicy server would have slowness send buffer would fill
- I/O stayed in queue for 2 seconds each rewind and may take seconds or tens of seconds to process
- The queue would never clear because of each 2 second rewind, even if the FPolicy server had a temporary performance problem
- The only way this would stop is if I/O dropped to near zero periods or manual intervention to disable FPolicy
- The 2 second window created a loop:
- ONTAP 9.8's much smaller EAGAIN rewind window allows the queue to clear when the FPolicy server starts responding
- The following conditions need to be met to consider increasing send buffer size:
- Latency must be seen on volumes
- The FPolicy vendor has seen performance is fine
- The network between ONTAP and the FPolicy server is clean and responsive as determined in a packet trace
- NetApp Support may recommend to increase outside of the above conditions in specific cases
- If none of above are true, please follow the guidance in High sustained latency after ONTAP upgrade to 9.8 and above due to FPolicy EAGAIN