Message ID | CAK2gM1vmua3z7ZQtudyeTQbxivw0qxkGt5K2kJQFJN9UsQL6CQ@mail.gmail.com (mailing list archive) |
---|---|
State | Not Applicable, archived |
Delegated to: | christophe varoqui |
Headers | show |
Arnon, On Tue, 2017-10-17 at 10:55 +0300, Arnon Yaari wrote: > Based on the manufacturer documentation: > https://support.infinidat.com/hc/en-us/articles/202319222 > > INFINIDAT recommends round-robin path selector using > a different path per IO. Timeout and path recovery values > are adjusted for error-free hot upgrade scenarios. > > Signed-off-by: Arnon Yaari <arnony@infinidat.com> > + .fast_io_fail = 15, > + .dev_loss = 15, I apologize for coming back to this after more than a year. I've been wondering about your dev_loss recommendation. What is the rationale for setting dev_loss and fast_io_fail to the same value, which is straight against the general recommendation? And what is the reason for the aggressively low dev_loss value anyway? Device loss and re-discovery is much more complex to handle for both the kernel and multipathd than failure/reinstantiation. You are the only vendor who recommends setting dev_loss less than the default of 600s. Could you share your reasoning please? Regards, Martin
On 11/7/18 11:41 AM, Martin Wilck wrote: > I apologize for coming back to this after more than a year. > I've been wondering about your dev_loss recommendation. > > What is the rationale for setting dev_loss and fast_io_fail to the same > value, which is straight against the general recommendation? And what > is the reason for the aggressively low dev_loss value anyway? Device > loss and re-discovery is much more complex to handle for both the > kernel and multipathd than failure/reinstantiation. You are the only > vendor who recommends setting dev_loss less than the default of 600s. > > Could you share your reasoning please? IMO, any change of a standard value should be documented. -- dm-devel mailing list dm-devel@redhat.com https://www.redhat.com/mailman/listinfo/dm-devel
Xose, On Wed, 2018-12-12 at 17:58 +0100, Xose Vazquez Perez wrote: > On 11/7/18 11:41 AM, Martin Wilck wrote: > > > I apologize for coming back to this after more than a year. > > I've been wondering about your dev_loss recommendation. > > > > What is the rationale for setting dev_loss and fast_io_fail to the > > same > > value, which is straight against the general recommendation? And > > what > > is the reason for the aggressively low dev_loss value anyway? > > Device > > loss and re-discovery is much more complex to handle for both the > > kernel and multipathd than failure/reinstantiation. You are the > > only > > vendor who recommends setting dev_loss less than the default of > > 600s. > > > > Could you share your reasoning please? > IMO, any change of a standard value should be documented. I agree. Have you looked at the past conversation (August 2017)? You challenged Arnon for explanations, but he came up with just a general statement ("Timeout and path recovery values are adjusted for error- free hot upgrade scenarios."), no detailed explanations. In particular, no rationale was given for the unusually aggressive dev_loss setting. That's the point of my inquiry. Martin
diff --git a/libmultipath/hwtable.c b/libmultipath/hwtable.c index 9e14ec1e..aa0ced53 100644 --- a/libmultipath/hwtable.c +++ b/libmultipath/hwtable.c @@ -964,13 +964,24 @@ static struct hwentry default_hw[] = { }, /* * Infinidat + * + * Maintainer: Arnon Yaari + * Mail: arnony@infinidat.com */ { .vendor = "NFINIDAT", .product = "InfiniBox", .pgpolicy = GROUP_BY_PRIO, - .pgfailback = -FAILBACK_IMMEDIATE, + .pgfailback = 30, .prio_name = PRIO_ALUA, + .selector = "round-robin 0", + .rr_weight = RR_WEIGHT_PRIO, + .no_path_retry = NO_PATH_RETRY_FAIL, + .minio = 1, + .minio_rq = 1, + .flush_on_last_del = FLUSH_ENABLED, + .fast_io_fail = 15, + .dev_loss = 15, }, /* * Nimble Storage
Based on the manufacturer documentation: https://support.infinidat.com/hc/en-us/articles/202319222 INFINIDAT recommends round-robin path selector using a different path per IO. Timeout and path recovery values are adjusted for error-free hot upgrade scenarios. Signed-off-by: Arnon Yaari <arnony@infinidat.com> --- Changes in v2: - Added maintainer information - Removed default values Changes in v3: - Changed patch message libmultipath/hwtable.c | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) -- 2.11.0 -- dm-devel mailing list dm-devel@redhat.com https://www.redhat.com/mailman/listinfo/dm-devel