[tor-dev] Dealing with critical sbws tickets
juga at riseup.net
Tue Jun 4 05:29:00 UTC 2019
> Hi juga,
> I read your meeting notes from this week's network team meeting:
> Week of 05/20 (planned)
> - Add Tor version to the bandwidth file (#30196)
> Week of 05/20 (actual)
> Week of 06/03 (plan)
> - Continue with #30406: Refactor header constants in sbws to
> use Stem's one
> For the next few weeks, can you focus on fixing critical sbws bugs,
> and helping with authority deployments?
Yes, i'll do my best with the little time i've to continue with sbws.
> Here's what I think we could do:
> I would like us to deploy sbws to 3/6 bandwidth authorities some time
> in June. We can do this deployment as soon as another directory
> authority operator is ready.
There's another authority operator ready. If you think we don't need to
fix any bug before a 3rd directory authority runs sbws, i can tell them
to start running sbws.
> To deploy more than 3 sbws instances, we need to fix these critical sbws
> We need sbws to generate bandwidth lines for all relays with results,
> even if they are not Running in the sbws tor client's current consensus.
> We need sbws to use MaxAdvertisedBandwidth from the latest descriptors:
> We also need to look for any more critical bugs in sbws. Here are some
> ways we can check for bugs:
> We need to check if all sbws instances exclude some relays, to help us find
> any more bugs in sbws:
> 90% of sbws measurement attempts fail. But these are internal errors, not
> network errors. So it looks like sbws has a relay selection bug:
> After we do these tasks, we can deploy sbws to 4 bandwidth authorities.
> What do you think?
I'll look at all these bugs more in detail.
How many directory authorities would we like to be running sbws (after
those bugs are fixed) by which date?.
BTW, longclaw's sbws did not have network for ~1 days (which for sure
has affected to some metrics), i should have documented that somewhere,
not sure there's a better place for that than trac.
More information about the tor-dev