Tor under DDoS, seen from the Bitcoin Network

Dennis Trautwein

Introduction

As of the 17th of October the ProbeLab team collects metrics about the Bitcoin network. The metrics comprise general topology information (e.g., where nodes reside geographically) but also specific data about nodes like their sync state and feature adoption.

On the 29th of October, Bitcoin nodes reachable as Tor onion services dropped by a third in our network data. Tor's own metrics show a large spike in relay clients over the same days which is apparently considered a DDoS attack on the Tor network. Nodes on IPv4, IPv6 and I2P did not change.

Quick recap of what we measure

Our flagship crawler, Nebula, crawls the Bitcoin P2P network about every six hours. It collects node addresses from peers and then dials each address. Onion addresses are dialed through a set of local Tor proxies. A node counts as connected when the dial succeeds.

What happened on the 29th of October

Metric 10-28 18:00 crawl 10-29 00:00 crawl
Onion peers connected 13,627 9,071
Onion peers dialed 28,845 28,733

On the 29th of October the number of peers we could connect to via Tor dropped from ~13.6k to 9.1k while the peers we tried to dial stayed almost constant. The step happened between the above two crawls, a drop of about 33%.

Onion peers dialed vs connected per 6h bucket, 10-20 to 11-15

Looking at the other available transports IPv4, IPv6 and I2P we see that these three stay within a few percent through the drop. Only Tor changed.

Peers connected by transport per 6h bucket, 10-20 to 11-15

Errors on onion dials

We compared failed onion dials in the 12 hours before 20:00 on 10-28 with the 12 hours after.

Error Before After
tor_onion_service_not_found 23,472 32,855
tor_onion_service_connect_failed 679 2,659
tor_stream_closed 2,523 769
tor_proxy_error 1,841 1,859

tor_proxy_error is flat, so the local Tor proxy was not the problem. Connect failures rose by about 4x.

Comparison with Tor Metrics

Tor Metrics estimates relay clients from directory requests. The estimate was about 3M per day until the 27th of October 2026. It then rose to 23.4M on the 28th and 64.5M on the 29th, stayed between 47M and 57M through the 2nd of November, and was back to about 3M by the sixth.

The Tor Project released 0.4.9.17 on the 13th of November as a response to an ongoing DDoS that is overwhelming relays with client circuits. The release note does not give a start date. The timing of the client spike matches our drop.

Tor relay users (log scale) and onion peers connected. Tor Metrics data ends 11-13.

Who came back

We continued to check who of the peers that were evicted from the Bitcoin network actually came back. We took the onion peers connected on 28th of October before 18:00 as a baseline.

Group Peers
Baseline 14,525
Still connected on 10-29 9,498
Lost on 10-29 5,027
Not in baseline, connected on 10-29 361

Of the 5,027 lost peers:

Status on 13th + 14th of November Peers
Connected again 2,126 (42%)
Still dialed, not connecting 2,898 (58%)
No longer in the crawl 3

Only 3 of the lost peers disappeared from the address set.

On 13th + 14th of November, 12.5k onion peers connected. 1.7k of them were not in the baseline. Daily onion peers that we could connect to rose from 9.9k on October 29th to 12.2k on November 14th, against 14.7k before the drop.

The recovery has one step. Onion peers connected went from 9.3k to 10.5k around November 5th. In the lost cohort, connected peers went from 361 to 1.5 during the same time.

Daily status of the 5,027 peers lost on 10-29

What we do not know

  • Why the failures moved from not_found to connect_failed. We have not confirmed whether hidden-service directories, introduction points or rendezvous circuits were the bottleneck. Tor-side data would be needed.
  • Whether the 2,898 peers still failing are affected by the same problem, offline, or rotated their onion address.
  • Our crawls run every 6 hours, so the start time is known only to within a crawl interval.

Conclusion

On the 29th of October 2026, onion-reachable Bitcoin nodes fell by about a third in our collected metrics while the number of onion addresses we dialed stayed constant. IPv4, IPv6 and I2P peers did not change. The drop starts the day the Tor Metrics relay-user estimate rises from about 3M to 64.5M. Bitcoin node recovery stepped on November 5th, while simultaneously the Tor relay-user estimate fell from 10.8M to 4.0M. 2.1k of the 5k lost Bitcoin peers (42%) have reconnected since. The timing of our drop and recovery matches the Tor relay-client spike. We have no Tor-side data on which part of the hidden-service path failed, and 2.9k of the lost peers still do not connect which requires further investigation.

Share this post

Related Posts