[rrd4j] Very poor performance on resource contrained systems

(Already discussed in https://github.com/openhab/openhab-addons/pull/8866#issuecomment-719944078)

I have noticed a huge performance degradation on the retrieval of data persisted in rrd4j on a PINE A64 2GB in recent snapshots (#1998), when the same operation on the same data used to be fast on the M1 release.

I’m attaching an example of RRD file where this can be reproduced: ElectricityMeter_Current (Number:Current) FGDW002WindowSensor_Temperature (Number:Temperature) rrd-examples.zip

The performance is as follows - these are exactly the same requests except in the second case the rrd4j persistence addon was uninstalled and replaced with the .jar from the Milestone 1 release:

3.0.0-SNAPSHOT (distribution version)

time curl -o /dev/null http://openhab:8080/rest/persistence/items/ElectricityMeter_Current?starttime=2020-10-31T10%3A47%3A36.624Z
real    0m2.996s   < !!!
user    0m0.030s
sys     0m0.012s

time curl -o /dev/null http://openhab:8080/rest/persistence/items/FGDW002WindowSensor_Temperature?starttime=2020-10-31T10%3A47%3A36.624Z
real    0m0.618s   < !!
user    0m0.031s
sys     0m0.013s

3.0.0.M1

time curl -o /dev/null http://openhab:8080/rest/persistence/items/ElectricityMeter_Current?starttime=2020-10-31T10%3A47%3A36.624Z
real    0m0.071s
user    0m0.023s
sys     0m0.018s

time curl -o /dev/null http://openhab:8080/rest/persistence/items/FGDW002WindowSensor_Temperature?starttime=2020-10-31T10%3A47%3A36.624Z
real    0m0.067s
user    0m0.019s
sys     0m0.021s

[wborn]

The bad performance probably has more impact because it causes issues all the time whereas the restore on startup behavior only occurs on startup… which occurs less often. WDYT?

[kaikreuzer]

it does not appear logical.

Ok, I think I found a logical explanation and fixed it with https://github.com/openhab/openhab-addons/pull/8938.