search

steemit / hivemind

Developer-friendly microservice powering social networks on the Steem blockchain.
MIT License
73 stars 66 forks source link

add timing log; add new where condition only get 7 days posts #336

Open ety001 opened 1 week ago

ety001 commented 1 week ago
only-dev-time commented 1 week ago

I'll run my tester over it for a day to see how long the response times are compared to my test in August.

Edit after testing: The most important results in comparison: no. method and parameter description avg response time old (ms) avg response time new (ms) Improvment (ms) %
1 bridge.get_ranked_posts, sort trending, max limit 100, tag my, observer remlaps 21796 548 -21248 -97%
2 bridge.get_ranked_posts, sort created, max limit 100, tag my, observer remlaps 825 16225 +15400 +1867%
3 bridge.get_ranked_posts, sort trending, max limit 100, tag my, observer barski (following 16T) timeout 786 no more timeout -
4 bridge.get_ranked_posts, sort created, max limit 100, tag my, observer barski (following 16T) timeout timeout - -
5 bridge.get_ranked_posts, sort trending, max limit 100, tag hive-185836 = community trends timeout 681 no more timeout -
6 bridge.get_ranked_posts, sort created, max limit 100, tag hive-185836 = community trends 25863 timeout - -

Brief description of the test: I sent various queries from the bridge_api every hour for one day. The queries are sent to my Hivemind test server. This excludes influences from other queries. The response time in min, max and average was then calculated for all responses.

The improvement in the requests with sort=trending (no. 1, 3, 5) can be seen very clearly. So the change had the desired effect here.

What surprises me, however, is the worsening of the requests with sort=created (no. 2, 4, 6). The change should have no effect at all on sort=created. At first I assumed network influences, but since this affects all created requests, I would exclude this.

Can anyone confirm this?

How ends your tests on logging level, @ety001?

10.09.2024: Update after repeating the test: I measured the response times again with the code before the change and after the change. The test was performed on the same server (with address localhost). Network problems can therefore be ruled out.