Description
This is the continuation of the issue #5764 . The machine I use and database I use is the same.
I have upgraded influxdb to 0.12 from debian (0.12.0-1). Now the start takes slightly over 2 hours. 33% increase is not bad, but still unacceptable to have downtime of 2 hours just to start. My dream is to replace MySQL server with InfluxDB, because I see InfluxDB is better to store my timeline data. MySQL starts within several seconds (I will keep memory consumption comparison for separate thread).
Can we improve InfluxDB to beat MySQL? :) I have my feeling that the long startup (and memory consumption also) is the price of being "schemaless", but in fact there is a schema that is kept in the memory. Loading that schema at startup takes long time and a lot of RAM (for certain cases). I might be wrong with my assumptions, sorry for being overconfident.
I've looked into logs, seems there's pretty long shard loads happening, I've grepped from log:
tsm1] 2016/04/06 21:43:56 /var/lib/influxdb/data/Loadosophia/default/512 database index loaded in 11m2.839406822s
[tsm1] 2016/04/06 21:47:00 /var/lib/influxdb/data/Loadosophia/default/513 database index loaded in 14m7.359681031s
[tsm1] 2016/04/06 21:54:48 /var/lib/influxdb/data/Loadosophia/default/525 database index loaded in 21m12.196889089s
[[tsm1] 2016/04/06 22:11:10 /var/lib/influxdb/data/Loadosophia/default/524 database index loaded in 27m14.029874898s
[tsm1] 2016/04/06 22:45:11 /var/lib/influxdb/data/Loadosophia/default/523 database index loaded in 1h12m5.557480759s
[tsm1] 2016/04/06 23:41:10 /var/lib/influxdb/data/Loadosophia/default/527 database index loaded in 1h54m9.795156493s
Attaching full startup log.
influxd.20160407.txt