[Merged by Bors] - collect latency and failure from every request and adjust latency based on response size #5601
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #5601 +/- ##
=========================================
+ Coverage 79.6% 79.8% +0.1%
=========================================
Files 270 270
Lines 27186 27232 +46
=========================================
+ Hits 21667 21733 +66
+ Misses 3973 3959 -14
+ Partials 1546 1540 -6 ☔ View full report in Codecov by Sentry. |
dshulyak
changed the title
|
@ivan4th could you please review the change, i think it may somewhat conflict with your streaming patch. if you have suggestion how to make it less conflicting, i can adjust. i will take a look later at streaming pr |
|
bors merge |
…ed on response size (#5601) closes: #5265 previous implementation was incomplete and had two flaws. it will not account for the difference in response size, larger payloads will naturally take more time to complete. it would skew peer selection logic towards peers that are slower to respond. the other flaw was to use only hs/1 protocol for measuing latency, often peers are blocked on ax/1 (where we get a collection of ids, which is large and slow) and lack of prioritization there may cause inefficient retries.
|
@dshulyak in #5562, an additional set of streaming fetcher methods are added which use new |
|
Build failed: |
|
bors merge |
…ed on response size (#5601) closes: #5265 previous implementation was incomplete and had two flaws. it will not account for the difference in response size, larger payloads will naturally take more time to complete. it would skew peer selection logic towards peers that are slower to respond. the other flaw was to use only hs/1 protocol for measuing latency, often peers are blocked on ax/1 (where we get a collection of ids, which is large and slow) and lack of prioritization there may cause inefficient retries.
|
Pull request successfully merged into develop. Build succeeded:
|
spacemesh-bors
bot
changed the title
closes: #5265
previous implementation was incomplete and had two flaws. it will not account for the difference in response size, larger payloads will naturally take more time to complete. it would skew peer selection logic towards peers that are slower to respond.
the other flaw was to use only hs/1 protocol for measuing latency, often peers are blocked on ax/1 (where we get a collection of ids, which is large and slow) and lack of prioritization there may cause inefficient retries.