Closed jonny-rimek closed 3 years ago
cost was 90€ down from 230, mainly cus of decreased lambda size, somehow I had 32GB more logs, mustve left debug on somewhere^^
Was only 1 function that had debug on, and according to log group no where near the ~40GB. should keep an eye on that
Further reduced max concurrent invocations to 10convert lambdas and 2x 5 get summary lambdas, no more complete failure. all throttles are retried and resolved evantually
The support increased the timestream query limit and I did another round of load tests at 200concurrency for convert, and 2x 100 for the query lambdas. The results are very promising, I got a few throttles, but overall no messages in dlq.
I only sent ~7,5k events instead of the 15k I usually do.
As a side effect of the increased limit the query lambdas are also faster
the write cost of the old load test with 7,5k event 22€ with 44,6gb written
7,5k events - common attributes
atm it looks like the the write cost of the load test was ~7€ which would be around 70% saving compared to before the common attribute. for some reason i have 550GB data scanned by queries, but each query is only 10MB, would be around 55k queries, seems too much. need to keep an eye on it.
the rest of the bill is reasonable, lambda is +4€ but part of it was in the free tier irrc
Another round of loadtesting. I'm now saving all combat logs and upload a 430MB file with 4 logs (1 deplete). The file is processed 1300 times. Important note: I doubled the convert lambda memory, but the writes are now concurrent, hence it's way faster at the same time the file is 10x bigger, which results in the same duration as before
This is the cost distribution before.
metrics:
xray
result:
unfortunately my first screenshot didn't capture timestream cost, the most important item :s lambda cost was around 2€
moved spell and caster id both to common attribute to reduce write cost in timestream
result:
I'm still waiting on my limit increase for timestream in the new accounts and i'll do the last load test once I refactored the upload process to a step function
the limit was increased in the prod account, another test after sfn refactor
Completed another round of load testing.
the only problem was query throttling in timestream, everything else worked perfectly. cost is not yet updated.