Having the end of the GlobalWindow be max - 1 day seems rather arbitrary. If it really needs to be less than MAX_TIMESTAMP, how about its immediate predecessor? (Or is there ambiguity about milliseconds vs. microseconds to deal with here? And even then, 2^63 - 1 micros is not likely representable as an even number of milliseconds...) Or is the concern that windowing from global windows -> fixed windows would overflow? In which case we'd still overflow for days (not unreasonable) and maybe should choose something much larger.
I think it'd be worth doing at least a cursory exploration into what it would take to "fix" Java. If that's too hard, at least we should make this a proto value (in standard windows perhaps)? [ Full content available at: https://github.com/apache/beam/pull/6315 ] This message was relayed via gitbox.apache.org for [email protected]
