Robert Milkowski wrote: > On 09/02/2010 14:04, Sambit Nayak wrote: >> Hi Robert, >> >> In case you are still at this problem and want to look at older CMM >> trace messages : >> >> cmm_dbg_buf is the CMM trace buffer, and it wraps around as you >> already noted. >> The buffer structure holds the previous snapshot of the trace buffer >> (before the wraparound), apart from the current view after the >> wraparound. >> Try checking : >> > *(cmm_dbg_buf+8)+1/s <--- gives you the older part, before the >> wraparound >> instead of >> > *cmm_dbg_buf/s <--- gives you the current buffer after the wraparound >> >> If the messages you are looking for are even older, then you can't >> see them in the trace buffer. :( >> >> > > > Thank you! This might be handy one day. > Unfortunately it's already too late... > > btw: is there a way to get a date of events in cmm dbg buffer? is it > the tm fields? (but it's not a unix timestamp...)
Yes, the only time visible in the CMM trace messages is the timestamp (tm field) which is derived from the high resolution timer. It is in 10 microsecs units, and wraps around every 10000 seconds. Look at the comments in dbg_print_buf::dbprintf_va() in the code file usr/src/common/cl/util/dbg_printf.cc : // The system prepends the thread id and high resolution time // (nanoseconds are dropped and so are the upper digits) // to the specified string. // The unit for timestamp is 10 microseconds. // It wraps around every 10000 seconds. // Ex: gethrtime() = X ns = X/1000 us = X/10000 10 micro sec. Regards, Sambit > > However I still haven't found the root-cause and because the > corruption happend also to some data within ZFS pool... anyway despite > writing much more data and running zfs scrub several times no new > corruption happened which indicates it was a temporary problem - which > is really bad as we can't be sure if/when it will happen again. > > The only suspicious thing I can see is er_other_discard=17 on a > brocade switch on a port where one of the nodes is connected to - but > I can't see any other errors on the node itself. > >
