On Wed, Oct 01, 2025 at 10:47:56AM +0000, zwiebelrouter via tor-relays wrote:
I am actually experiencing the same problem, but with a OBFS4 bridge I've set up over a year ago on FreeBSD.
The bridge keeps getting marked as down for a few hours and I am receiving Tor weather down mail each time, even though it is always online and reachable (ORPort + OBFS4).
Hi! Thanks for running a bridge. I think your situation is different than the other person's issue. Your bridge gets a *lot* of use, presumably because many people have learned about it and use it as their bridge. So it makes sense that it continues to get use even when it's marked as down by rdsys and bridgestrap -- the users keep using it independent of whether it is marked down or given out to further users at that moment. One possible explanation is that it is overloaded to the point that it is inconsistent at receiving new incoming connections. That is, most of the time bridgestrap can successfully connect to it, but sometimes bridgestrap fails because you are out of some resource (file descriptors, socket accept queue, bandwidth, etc) at that moment. Another possible explanation is that some bug in the anti-censorship team toolchain (rdsys, bridgestrap, maybe the metrics side) is making it be marked down when it shouldn't be.
Status: https://bridges.torproject.org/status?id=159DAE6BC567CAE6F87281077518B6593C4...
Some days ago I tried this url and it said your bridge is working. But I just tried going there this moment and it says * obfs4 IPv4: dysfunctional Error: timed out waiting for bridge descriptor Last tested: 2025-10-04 18:15:42.513154413 +0000 UTC (2h47m17.286882518s ago) * obfs4 IPv6: dysfunctional Error: timed out waiting for bridge descriptor Last tested: 2025-10-04 18:15:42.513154413 +0000 UTC (2h47m17.286890098s ago) Yet I can still connect when I tried just now (albeit almost 3 hours after bridgestrap tried and failed so that doesn't say that much). So my guess is more toward 'inconsistently unreachable, perhaps because overloaded' rather than some toolchain bug. --Roger