Hello Ilya.
Thank you so much for your answer.
I tried your solution in the same scenario:
-       PARTITIONED cache, SQL indexed by key and by a field (type long, the 
last
key field) descending
-       checkpoint frequency restored to default
-       4000 msg/sec rate (fixed)
I set the environment variable IGNITE_MAX_INDEX_PAYLOAD_SIZE to 66:
(WARN 59726 --- [0-#17%Datafeed%] o.a.i.i.p.query.h2.database.H2TreeIndex  :
<DF_CMF_QUOTE> Indexed columns of a row cannot be fully inlined into index
what may lead to slowdown due to additional data page reads, increase index
inline size if needed (set system property IGNITE_MAX_INDEX_PAYLOAD_SIZE
with recommended size (be aware it will be used by default for all indexes
without explicit inline size)) [cacheName=IGN_DF_CMF_QUOTE,
tableName=DF_CMF_QUOTE, idxName=_key_PK, idxCols=(_KEY), idxType=PRIMARY
KEY, curSize=10, recommendedInlineSize=66])
in all cluster nodes and I ran the test again.
We had to stop the test after 3 milion messages and what I could notice was
that the checkpoints collected a number of pages that grew over and over.
This increment grew much faster than in previous tests
(IGNITE_MAX_INDEX_PAYLOAD_SIZE not set). 
So, I couldn't detect the performance degradation (because I had to stop the
test necessarily) but the increasing number of pages in every checkpoints
let me expect it will happen.

I attach the server logs of this test.

log_ignite_190227_setIndexPayload.gz
<http://apache-ignite-users.70518.x6.nabble.com/file/t2315/log_ignite_190227_setIndexPayload.gz>
  


What do you think about ?

Thanks a lot

Regards




--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Reply via email to