Messages in this thread | | | Date | Sun, 14 Apr 2024 11:34:58 +0300 | Subject | Re: [PATCH v5 0/6] nvme-fabrics: short-circuit connect retries | From | Sagi Grimberg <> |
| |
On 12/04/2024 18:24, Keith Busch wrote: > On Fri, Apr 12, 2024 at 09:24:04AM +0200, Daniel Wagner wrote: >> On Thu, Apr 11, 2024 at 06:35:25PM -0600, Keith Busch wrote: >>> On Tue, Apr 09, 2024 at 11:35:04AM +0200, Daniel Wagner wrote: >>>> The first patch returns only kernel error codes now and avoids overwriting error >>>> codes later. Thje newly introduced helper for deciding if a reconnect should be >>>> attempted is the only place where we have the logic (and documentation). >>>> >>>> On the target side I've separate the nvme status from the dhchap status handling >>>> which made it a bit clearer. I was tempted to refactor the code in >>>> nvmet_execute_auth_send to avoid hitting the 80 chars limit but didn't came up >>>> with something nice yet. So let's keep this change at a minimum before any >>>> refactoring attempts. >>>> >>>> I've tested with blktests and also an real hardware for nvme-fc. >>> Thanks, series applied to nvme-6.9. >> Thanks! I have an updated version here which addresses some of Sagi's >> feedback, e.g. using only one helper function. Sorry I didn't send out >> it earlier, I got a bit side tracked in testing because of the 'funky' >> results with RDMA. >> >> Do you want me to send a complete fresh series or patches on top of this >> series? I'm fine either way. > Oh sorry, I didn't notice the discussion carried on after the "review" > tag. Please send me the update, I'll force push.
I think that what we want is to have a special handling in the very specific connect failure when the host/ctrl credentials mismatch, because out of all the reasons to a connection failure, this _could_ be transient.
| |