Storage Server Version Triage #1007
Conversation
neuroscr
commented
Mar 25, 2020
•
neuroscr
commented
- builds a pool per version (and a reverse map)
- filters pools with 2.0.2 or greater for file server proxy
- mark as bad removes from random and version pools
- increased start up retries from 1s to 10s (as this PR adds 10-20s delay to proxy services while we get versions and sorts in the usable buckets)
- takes about 10mins to get versions from 980 sndoes
….2 or newer versions
…sionPools now, lint
… 1s, randomize list of snodes, only report status every 10%
neuroscr
requested review from
msgmaxim and
Mikunj
and removed request for
msgmaxim
What would this mean in terms of startup time and receiving messages?
Wouldn't it be better to query snodes until we have X amount that are of a certain version, and keep doing it in the background? |
Only should really affect proxy requests (RSS, version check, attachments, device mapping fetching)
The version fetching is backgrounded for the most part. |
|
Notes for V2:
|
…able() make handle edge removal cases and optimize snode lookup
|
Refined patch to only affect file upload (attachments/avatars) proxy requests allowing RSS, version check, and device mapping fetching to proceed normally since their content is usually under 1mb. Also handled some edge cases in the markRandomSNodeBad when we haven't got it's version yet, so we retry, so we can mark it bad if/when we do get it. |
| await this.refreshRandomPool(); | ||
| } | ||
| if (this.randomSnodePool.length === 0) { | ||
| throw new window.textsecure.SeedNodeError('Invalid seed node response'); |
There was a problem hiding this comment.
Should we throw this in refreshRandomPool() instead?
There was a problem hiding this comment.
Not sure, @Mikunj do you know if this is an issue that we throw inside a promise instead of the outer function?
There was a problem hiding this comment.
Throwing inside async should reject the promise and awaiting on rejected promise should just re-throw the original exception.
There was a problem hiding this comment.
which we catch on line 483 and rethrow
|
I really hate how this floods the log with |
