Output should be identical, but I haven't tested this.
AFAICT the docs don't explicitly guarantee that the output column order will match the order supplied to query-gpu, but it would be weird to offer noheader otherwise.
It may be necessary to manually insert empty columns in some collectors, but in this case
we're already assuming that if we get even one line for a given GPU it includes all our measurements, and
we must get at least one line for each GPU for it to be saved in gpu_indices anyway,
so it's safe to fill them in as they're encountered.
Is there a particular reason to expect lines of less than two fields? (Or is that check just to prevent crashes? Because it doesn't, if the field count is greater than 2 but less than len(measurements)+2. Also, if so, should probably emit a warning once we get logging fixed.)
Output should be identical, but I haven't tested this.
AFAICT the docs don't explicitly guarantee that the output column order will match the order supplied to
query-gpu
, but it would be weird to offernoheader
otherwise.It may be necessary to manually insert empty columns in some collectors, but in this case
gpu_indices
anyway,so it's safe to fill them in as they're encountered.
Is there a particular reason to expect lines of less than two fields? (Or is that check just to prevent crashes? Because it doesn't, if the field count is greater than
2
but less thanlen(measurements)+2
. Also, if so, should probably emit a warning once we get logging fixed.)