On 11/28/2016 12:26 PM, Erick Erickson wrote:
Well, such checks could be put in, but they don't get past the basic problem.
And all this masks your real problem; you didn't have enough disk
space to optimize in the first place. Even during regular indexing w/o
optimizing, Lucene segment merging can always theoretically merge all
your segments at once. Therefore you always need at _least_ as much
free space on your disks as all your indexes occupy to be sure you
won't hit a disk-full problem. The rest would be band-aids. Although I
suppose refusing to even start if there wasn't enough free disk space
isn't a bad idea, it's not foolproof though
If such a "warning" feature is added, it's not "foolproof" that I would
expect, and I wouldn't expect it be able to "predict" usage caused by
events happening after it passes a basic initial check. I am just
thinking a basic up-front check that indicates "it just ain't happening"
might be useful.
So.. how does one handle needing all this "free space" between major
index updates when one gets charged by the GB for allocated space
without regard to actual storage usage?
