Before I file an issue on this, I wanted to bring it up here, so I can
see if there's something I'm overlooking.
Distributed MLT is very very slow for me. I can make it work, but a
QTime of one to two minutes in production isn't acceptable. Sending a
non-distributed MLT request directly to a large shard takes about 1.5
seconds. There are six large cold shards and one tiny hot shard.
I used my dev server to gather some logs. This server is considerably
less powerful than my production servers, but has exactly the same data.
It's running a 4.5 snapshot with the patch from SOLR-5125. Unlike my
production servers, the dev server takes over four minutes for the
distributed MLT request. Slightly redacted logfile at this URL:
https://dl.dropboxusercontent.com/u/97770508/slow-mlt.log
After I ran the query that you can see in the logfile, I restarted Solr
on my dev server and ran one of the slow subrequests directly to a
shard. Here's the debugQuery timing section from that request. QTime
on it was 56506:
"QParser":"LuceneQParser",
"timing":{
"time":56504.0,
"prepare":{
"time":29.0,
"query":{
"time":29.0},
"facet":{
"time":0.0},
"mlt":{
"time":0.0},
"highlight":{
"time":0.0},
"stats":{
"time":0.0},
"spellcheck":{
"time":0.0},
"debug":{
"time":0.0}},
"process":{
"time":56475.0,
"query":{
"time":935.0},
"facet":{
"time":0.0},
"mlt":{
"time":55442.0},
"highlight":{
"time":0.0},
"stats":{
"time":0.0},
"spellcheck":{
"time":0.0},
"debug":{
"time":98.0}}}}}
Is there anything for me to do other than file an issue?
Thanks,
Shawn
