New conflux links every 30 sec when unused connections

changed milestone to %Tor: 0.4.9.x-freeze

changed milestone to %Tor: 0.4.8.x-post-stable

@mikeperry Hmmm so what cycles circuits every 30 seconds like that? ... Circuit expire build time function or sth? You have a theory on top of your head?

Yeah build time expiry is the only thing I can think of..

The real question is why does it keep doing this forever? What is actually preventing any circuit from completing, until activity resumes?

Why does resuming activity allow them to succeed? (And what does resuming activity mean in this case?)

DFLT_NONPRIMARY_GUARD_CONNECT_TIMEOUT from entrynodes.h is 15 seconds.. 2*15 is 30... Getting warmer...

I think I hit some version of this: is it possible that we're getting stuck in some kind of GUARD_WAIT state, where we don't like the guard for the current circuits, and let them die?

I am very much not understanding this GUARD_WAIT code, and it does look like that is just what happened to me. I am not sure how to cause it to happen, though. It wasn't just 30 minutes idle..

I think that a workaround for this issue would look something like:

diff --git a/src/core/or/circuitbuild.c b/src/core/or/circuitbuild.c
index d6e022e7fa..712319d4b3 100644
--- a/src/core/or/circuitbuild.c
+++ b/src/core/or/circuitbuild.c
@@ -1092,6 +1092,15 @@ circuit_build_no_more_hops(origin_circuit_t *circ)
   } else {
     r = entry_guard_succeeded(&circ->guard_state);
   }
+
+  /* Exempt conflux circuits from this wait-for-usable-guard madness:
+   * if only 1 leg is in this wait state, it will repeatedly get closed
+   * due to timeout, only to retry again and again.
+   */
+  if (circ->base_.purpose == CIRCUIT_PURPOSE_CONFLUX_UNLINKED) {
+    r = GUARD_USABLE_NOW;
+  }
+
   const int is_usable_for_streams = (r == GUARD_USABLE_NOW);
   if (r == GUARD_USABLE_NOW) {
     circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_OPEN);

but I am unable to regularly reproduce this issue. Simply waiting 30 minutes as @cypherpunks said does not do it for me. In my case, it has something to do with being disconnected or suspended.

This might also be an entirely different issue than @cypherpunks is reporting.

assigned to @mikeperry

added Doing label

guardmon.py

Here's a control port script to check circuit status.

On rare occasion, I end up stuck in GUARD_WAIT for a while, spamming circ attempts, but it eventually goes away. My assumption is that this is the bug?

This looks like a dup of the bug @arma just found and fixed: #40981 (closed). The fix was just merged: !838 (merged)

Whoops! Good find. Yes, based on the description this is exactly the bug that I fixed.

I'm going to close this one as a duplicate.

closed

Indeed same bug, can confirm that was fixed in 4.8.13. Don't forget to close the other issue #40981 (closed).

New conflux links every 30 sec when unused connections

Child items 0

Activity