Rob van der Heij wrote:
> With a "unique first" on OK/error you can at least produce the "run
> number" and prefix the server key with that. Use that to sort it on
> run&server and take first and last of each.
Aha, that's probably far more sensible than a separate pipeline for each
run or each server, since the whole point of the exercise is to keep the
file small. I'd still have to number the records if I care about
maintaining the order of records with identical timestamps.
The sipping pipeline for each run was what I was thinking of as I
accidentally hit "send" last night:
callpipe (end /) *: | totarget ... | notfirst: lookup autoadd ...
| instore reverse | outstore | not lookup autoadd ...
| instore reverse | outstore | first: merge ... | *:
/ notfirst: | buffer | first:
(where the ellipses after TOTARGET and MERGE could include the steps of
adding and removing record numbers)
¬R
