[tor-relays] relayor v26.2.0 is released - Goodbye "MyFamily"

Hi,

relayor v26.2.0 is released.

This release marks the end of "MyFamily" support in relayor.
Since all supported tor versions support the Happy Family design there is no reason to continue supporting the old "MyFamily" method to declare relay families.

relayor is an ansible role that helps you with automating tor relay operations

Changes since v26.1.0:
- remove MyFamily support: The last tor version that required MyFamily (0.4.8.x) in addition to Happy Family support reached end-of-life on 2026-06-01
- use absolute filter name to fix error "No filter named 'ipaddr'"

kind regards,
nusenu

···

--

_______________________________________________
tor-relays mailing list -- tor-relays@lists.torproject.org
To unsubscribe send an email to tor-relays-leave@lists.torproject.org

Hi nusenu,

I am not a fan of breaking security for clients not running the latest
supported release, while those clients continue to work on the network.
I have started a discussion among dirauth operators on how to deal with
relays who are actively harming outdated clients like that.

I appreciate your eagerness to push forward, but would argue the harm
caused by removing MyFamily across the board is greater than keeping it
around for some time. There are already network team plans to break
compatibility with an update which would actually mean old clients
refuse to function, which IMO is much better than breaking security
silently.

Thanks for your consideration
Sebastian

···

On 1. Jun 2026, at 20:19, nusenu via tor-relays <tor-relays@lists.torproject.org> wrote:

Hi,

relayor v26.2.0 is released.
Release relayor v26.2.0 · nusenu/ansible-relayor · GitHub

This release marks the end of "MyFamily" support in relayor.
Since all supported tor versions support the Happy Family design there is no reason to continue supporting the old "MyFamily" method to declare relay families.

relayor is an ansible role that helps you with automating tor relay operations
GitHub - nusenu/ansible-relayor: An Ansible Role for Tor Relay Operators · GitHub

Changes since v26.1.0:
- remove MyFamily support: The last tor version that required MyFamily (0.4.8.x) in addition to Happy Family support reached end-of-life on 2026-06-01
- use absolute filter name to fix error "No filter named 'ipaddr'"

kind regards,
nusenu

_______________________________________________
tor-relays mailing list -- tor-relays@lists.torproject.org
To unsubscribe send an email to tor-relays-leave@lists.torproject.org

Sebastian Hahn via tor-relays:

Hi,

relayor v26.2.0 is released.
Release relayor v26.2.0 · nusenu/ansible-relayor · GitHub

This release marks the end of "MyFamily" support in relayor.
Since all supported tor versions support the Happy Family design there is no reason to continue supporting the old "MyFamily" method to declare relay families.

relayor is an ansible role that helps you with automating tor relay operations
GitHub - nusenu/ansible-relayor: An Ansible Role for Tor Relay Operators · GitHub

Changes since v26.1.0:
- remove MyFamily support: The last tor version that required MyFamily (0.4.8.x) in addition to Happy Family support reached end-of-life on 2026-06-01
- use absolute filter name to fix error "No filter named 'ipaddr'"

kind regards,
nusenu

Hi nusenu,

I am not a fan of breaking security for clients not running the latest
supported release, while those clients continue to work on the network.
I have started a discussion among dirauth operators on how to deal with
relays who are actively harming outdated clients like that.

I appreciate your eagerness to push forward, but would argue the harm
caused by removing MyFamily across the board is greater than keeping it
around for some time. There are already network team plans to break
compatibility with an update which would actually mean old clients
refuse to function, which IMO is much better than breaking security
silently.

To add to that: there is a transition process as always when a Tor version goes EOL. We've been done that several times in the past and that should not be anything new. Moreover, as Sebastian indicated, we have this time a hard break for 0.4.8.x clients as well, which is publicly documented in Gitlab[1] and scheduled for 2026-09-01. By then 0.4.8.x relays will likely be blocked from the network as well. Thus, supporting MyFamily until then seems like a good choice to me to avoid at least confusion among relay operators, in particular as we link to your tool and talk about setting up MyFamily in the post-install section of our community portal.[2]

I'd suggest you reconsider just pulling the plug.

Georg

[1] Choose a date after which 0.4.8 clients do not have to work (#460) · Issues · The Tor Project / Network Health / Team · GitLab
[2] Tor Project | Relay Post-install and good practices

···

On 1. Jun 2026, at 20:19, nusenu via tor-relays <tor-relays@lists.torproject.org> wrote:

There is a time for this, but "now" is a poor choice.

When a release goes EOL, we have a period where we can work with other people and partners, who are in contact with us, about their use of Tor. During this time, we can hint to them at upgrading. Some of the reasons for upgrading are new features, safety, and security issues. We are not in a hurry at all here.

Getting them to upgrade is a goal we all should share, but screwing over their userbase shouldn't be.

Cheers,
Alex

···

On 01/06/2026 20.19, nusenu via tor-relays wrote:

- remove MyFamily support: The last tor version that required MyFamily (0.4.8.x) in addition to Happy Family support reached end-of-life on 2026-06-01

--
Alexander Hansen Færøy
_______________________________________________
tor-relays mailing list -- tor-relays@lists.torproject.org
To unsubscribe send an email to tor-relays-leave@lists.torproject.org

Hi Sebastian,

I have started a discussion among dirauth operators on how to deal with
relays who are actively harming outdated clients like that.

As someone who advocated for many years that the torproject and tor
directory authorities should take a stronger stance on relays without proper
family configuration I really appreciate if tor directory authorities would finally
be more active and take actions against such relays.

That would be unexpected and surprising but a very welcome and positive outcome.

Please do not limit the removal of relays to just relays putting end-of-life tor clients at risk but **also**
include relays that put currently supported tor clients at risk. MyFamily does not support super large groups as you know.

To avoid surprising relay operators I would suggest to officially announce that change in actually enforced policy
because operators are not used to it since many run relays in their current configuration since a long time
and directory authorities never did take any actions against them.

I've been in contact with one large operator that would be affected and they confirmed that they will update their configuration soon.

In OrNetStats I had(*) a dedicated section for relay groups that have an actual chance to perform end-to-end correlation attacks:

Operators are only listed if they actually have a chance to do end-to-end correlation attacks, that is:

- their guard and exit probability is > 0%
- they did not properly configure MyFamily
- they run in more than a single /16 network block

(*) unfortunatelly onionoo data quality is still not back to how it used to be before the CollecTor/collector-rs change on 2026-04-02,
therefore OrNetStats has not been updated since March 2026 because it depends on collector and onionoo data.

kind regards,
nusenu

···

--

_______________________________________________
tor-relays mailing list -- tor-relays@lists.torproject.org
To unsubscribe send an email to tor-relays-leave@lists.torproject.org

Georg Koppen via tor-relays:

Sebastian Hahn via tor-relays:

Hi,

relayor v26.2.0 is released.
Release relayor v26.2.0 · nusenu/ansible-relayor · GitHub

This release marks the end of "MyFamily" support in relayor.
Since all supported tor versions support the Happy Family design there is no reason to continue supporting the old "MyFamily" method to declare relay families.

relayor is an ansible role that helps you with automating tor relay operations
GitHub - nusenu/ansible-relayor: An Ansible Role for Tor Relay Operators · GitHub relay-operator

Changes since v26.1.0:
- remove MyFamily support: The last tor version that required MyFamily (0.4.8.x) in addition to Happy Family support reached end- of-life on 2026-06-01
- use absolute filter name to fix error "No filter named 'ipaddr'"

kind regards,
nusenu

Hi nusenu,

I am not a fan of breaking security for clients not running the latest
supported release, while those clients continue to work on the network.
I have started a discussion among dirauth operators on how to deal with
relays who are actively harming outdated clients like that.

I appreciate your eagerness to push forward, but would argue the harm
caused by removing MyFamily across the board is greater than keeping it
around for some time. There are already network team plans to break
compatibility with an update which would actually mean old clients
refuse to function, which IMO is much better than breaking security
silently.

To add to that: there is a transition process as always when a Tor version goes EOL. We've been done that several times in the past and that should not be anything new. Moreover, as Sebastian indicated, we have this time a hard break for 0.4.8.x clients as well, which is publicly documented in Gitlab[1] and scheduled for 2026-09-01. By then 0.4.8.x relays will likely be blocked from the network as well. Thus, supporting MyFamily until then seems like a good choice to me to avoid at least confusion among relay operators, in particular as we link to your tool and talk about setting up MyFamily in the post-install section of our community portal.[2]

I'd suggest you reconsider just pulling the plug.

@nusenu: is there any progress on that? We now had one dir-auth operator, folks from the network and network-health team weighting in and suggesting that your change was well-intentioned but rushed. It would be nice if you could update relayor accordingly.

Thanks,
Georg

···

On 1. Jun 2026, at 20:19, nusenu via tor-relays <tor- >>> relays@lists.torproject.org> wrote:

Georg

[1] Choose a date after which 0.4.8 clients do not have to work (#460) · Issues · The Tor Project / Network Health / Team · GitLab
[2] Tor Project | Relay Post-install and good practices

_______________________________________________
tor-relays mailing list -- tor-relays@lists.torproject.org
To unsubscribe send an email to tor-relays-leave@lists.torproject.org

TLDR: There appears to be a mismatch between the support policy [1] vs. actual expectations from torproject people,
it would be beneficial to clarify and update the policy and documentation to avoid similar situations
(different expectations and understandings of what EOL means in the context of the torproject)
in the future and so we and others can learn from it and hopefully prevent it from happening again.

Shortly after the release and feedback in this thread I added a note to the release information about delaying the upgrade to relayor v26.2.0
if the operator runs relays across more than one /16 IPv4 or /32 IPv6 network.

ahf wrote:

When a release goes EOL, we have a period where we can work with
other people and partners, who are in contact with us, about their
use of Tor. During this time, we can hint to them at upgrading. Some
of the reasons for upgrading are new features, safety, and security
issues. We are not in a hurry at all here.

It is clear now - after the emails in this thread - that some have expectations that go
beyond the end-of-life date of a tor release series.

1) The current support policy does not document the mentioned stages after the EOL date.
The current support policy [1] from Jul 19, 2022 describes these stages:

        -> alpha
         -> release candiate (rc)
          -> stable
           -> oldstable
            -> end-of-life

So according to that policy the last event in the lifecycle of a tor release series is the end-of-life date.
End-of-life releases are unsupported and "You are on your own.":

[1]:

After that, the release will be considered End of Life (EOL) as in
unsupported meaning that at any point after that state is set for the release
series, the directory authorities might reject them from the consensus.

Most notably [1] and [2] do not say anything about the mentioned additional stages beyond the end-of-life date,
in the specific case of 0.4.8 the stage between 2026-06-01 - 2026-09-01 and the stage after 2026-09-01.

2) Clarifying [1] with the mentioned expectations would be beneficial

I also created a gitlab issue for this under:

a) EOL meaning

Since the term "end-of-life" already has a certain meaning in the software world Software release life cycle - Wikipedia
I think it is best do not create a torproject specific meaning of "end-of-life" to avoid similar confusion in the future.

The "oldstable" description [1]:

Note that old stable are there to allow a transition period to the new stable
and generally not to be used for general use.

somewhat matches what ahf described ("During this time, we can hint to them at upgrading. Some of the reasons for upgrading are new features, safety, and security issues.")
to some degree.

b) Scope

I would also suggest to add scope information to the support policy to make the scope/roles of the policy clear (clients, onion services, relays, bridges, dir auths, ...)
because the current version includes wording that would suggest that it is only about relays by using terms like "the directory authorities might reject them from the consensus"
because clients are not in the consensus.
A pointer for arti's support policy would be good if it is not on the same page.

c) document non-mutual expectations

If the torproject expects relay operators to support tor client versions for a longer period of time
than the torproject it self supports them, then make that clear in the expectations for relay operators document [3] by pointing to [2].
An updated version of [2] could help relay operators find out the specific date until when they must support a specific tor series.
Currently it does not include such information (for example the 2026-09-01 date for 0.4.8.x).
Without explicit information many will believe it is fine to drop support as soon as the torproject declares them unsupported.

It is probably confusing for relay operators if they get removed from the tor network for using a end-of-life tor version
while also getting removed from the tor network for NOT supporting EOL version clients.

3) more proactive communications / announcements

On 2026-04-13 I asked about the EOL date [10] and suggested a blog post announcement, no reaction to the email was observed before the EOL date on 2026-06-01
but I noticed related gitlab activity after that email.

On my side:
- I will start proactive information for similar relayor changes but they were very rare in the past (1x in 10 years?).
- I might create a support policy for relayor - blocked until the torproject support policy changes are completed.
- I'll announce similar changes on my mastodon feed before they take place and are released.

kind regards,
nusenu

Some pointers to policies/documents, feel free to add more if you notice any relevant page missing:

[1] SupportPolicy · Wiki · The Tor Project / Core / Team · GitLab
[2] CoreTorReleases · Wiki · The Tor Project / Core / Team · GitLab
[3] Expectations for relay operators - The Tor Project - Policies

[4] Relay EOL policy · Wiki · The Tor Project / Network Health / Team · GitLab
[5] ReleasePolicy · Wiki · The Tor Project / Core / Team · GitLab
[6] TROVE · Wiki · The Tor Project / Core / Team · GitLab
[7] DeprecationTorPhases · Wiki · The Tor Project / Core / Team · GitLab
[8] Recommended Tor versions policy - The Tor Project - Policies
[9] Reject relays running unsupported Tor versions even if the series is not EOL yet (#149) · Issues · The Tor Project / Network Health / Team · GitLab
[10] [tor-relays] tor version 0.4.8.x end of life date? - tor-relays - lists.torproject.org

···

--

_______________________________________________
tor-relays mailing list -- tor-relays@lists.torproject.org
To unsubscribe send an email to tor-relays-leave@lists.torproject.org

While I support this argument, I wish to point out the hypocrisy of anyone
so involved in the Tor Project, which itself violated this policy early on and
has continued to do so for over 20 years, now making the argument at this late
date.
     When I first became aware of tor and began to use it, I was confused for
some time by its misuse of the term "alpha" releases. From the mid 1960s
through the early 1990s, during which time the vast majority of software
publicly available was written by hardware vendors, e.g., Burroughs, CDC, DEC,
Fujitsu/Facom, Harris, Honeywell, IBM, ICL, NCR. In the customer (a.k.a.
"user") communities, beta releases were known, but very few customer sites
chose to sign the vendors' contracts for the privilege of being on the
bleeding edge of development by being a beta test site. Those that did so,
however, were reputed to receive intensive support in analyzing bugs and
producing fixes for them. Alpha releases, by contrast, were basically out of
view outside the vendor corporations themselves and were basically under tight
wraps. It was occasionally possible to get some company representative to
admit that alpha releases did exist, but they wouldn't admit to having any
information about them except that they were strictly internal to those
corpations. Thus, after having always heard the terms used with these
understandings, it took me a while to understand that the Tor Project was yet
another victim of NIH Syndrome in its own small way. :slight_smile:
     Another example in a much larger way was the CSRG at U.C. Berkeley, who
apparently ignored much of the terminology previously invented elsewhere in the
development of support for virtual memory systems and instead invented their
own. All of the modern BSDs continue to use that vocabulary. One serious,
deficiency, IMO, of the CSRG's terminology was the lack of the term "page
frame" and the consequent substitution of the word "page", which results in
their use of "page" for two different, but very related, concepts. That can
lead to a great deal of confusion in some situations.
     Anyway, yes, please do publish clear definitions for the different phases
of the tor project's software support cycle (a.k.a. "life" cycle).

                                  Scott Bennett, Comm. ASMELG, CFIAG

···

nusenu via tor-relays <tor-relays@lists.torproject.org> wrote:

TLDR: There appears to be a mismatch between the support policy [1] vs. actual expectations from torproject people,
it would be beneficial to clarify and update the policy and documentation to avoid similar situations
(different expectations and understandings of what EOL means in the context of the torproject)
in the future and so we and others can learn from it and hopefully prevent it from happening again.

Shortly after the release and feedback in this thread I added a note to the release information about delaying the upgrade to relayor v26.2.0
if the operator runs relays across more than one /16 IPv4 or /32 IPv6 network.

ahf wrote:
> When a release goes EOL, we have a period where we can work with
> other people and partners, who are in contact with us, about their
> use of Tor. During this time, we can hint to them at upgrading. Some
> of the reasons for upgrading are new features, safety, and security
> issues. We are not in a hurry at all here.

It is clear now - after the emails in this thread - that some have expectations that go
beyond the end-of-life date of a tor release series.

1) The current support policy does not document the mentioned stages after the EOL date.
The current support policy [1] from Jul 19, 2022 describes these stages:

        -> alpha
         -> release candiate (rc)
          -> stable
           -> oldstable
            -> end-of-life

So according to that policy the last event in the lifecycle of a tor release series is the end-of-life date.
End-of-life releases are unsupported and "You are on your own.":

[1]:
> After that, the release will be considered End of Life (EOL) as in
> unsupported meaning that at any point after that state is set for the release
> series, the directory authorities might reject them from the consensus.

Most notably [1] and [2] do not say anything about the mentioned additional stages beyond the end-of-life date,
in the specific case of 0.4.8 the stage between 2026-06-01 - 2026-09-01 and the stage after 2026-09-01.

2) Clarifying [1] with the mentioned expectations would be beneficial

I also created a gitlab issue for this under:
support policy clarifications related to end-of-life (#32) · Issues · The Tor Project / Core / Team · GitLab

a) EOL meaning

Since the term "end-of-life" already has a certain meaning in the software world Software release life cycle - Wikipedia
I think it is best do not create a torproject specific meaning of "end-of-life" to avoid similar confusion in the future.

**********************************************************************
* Internet: bennett at sdf.org *xor* bennett at freeshell.org *
*--------------------------------------------------------------------*
* "A well regulated and disciplined militia, is at all times a good *
* objection to the introduction of that bane of all free governments *
* -- a standing army." *
* -- Gov. John Hancock, New York Journal, 28 January 1790 *
**********************************************************************
_______________________________________________
tor-relays mailing list -- tor-relays@lists.torproject.org
To unsubscribe send an email to tor-relays-leave@lists.torproject.org

Hi Nusenu,

I am reopening this thread because I want us to be on the same side here, but let me start by saying your feedback during the Happy Families work in C Tor was genuinely useful, and it made the feature easier to integrate with tools like yours.

A bit of context on why we are moving carefully here. Happy Families came out of the Walking Onions design work for Arti, where MyFamily does not scale. We backported it to C Tor because relay operators were hitting real pain: descriptor size limits as AuthDirMaxServersPerAddr went from 2 to 4 to 8, and the quadratic MyFamily configuration burden that ansible-relayor exists to smooth over.

To give you some context. What we are trying to avoid is breaking clients that are still out there and that we do not fully understand. That is why I would rather spend the next few months (just over 1 month really at this point) nudging downstream distributions toward 0.4.9.x than accelerate adoption. People have waited a long time for Happy Families, and a few more months of MyFamily pain is a trade I think most operators would accept.

On the framing: I do not see this as "relay operators vs. torproject.org". My ask is specifically to you as the maintainer of a widely used configuration tool, in the same way we would raise it with any tool that steers Tor toward behaviour we are recommending against.

Concretely, what I am asking is that you revert the 17-line diff and cut a release. A note on the GitHub tarball page will not reach people, since most users pull ansible-relayor as a Git submodule or from Ansible Galaxy. How you handle support and releases beyond that is entirely your call.

Would you be open to that?

-hiro

···

_______________________________________________
tor-relays mailing list -- tor-relays@lists.torproject.org
To unsubscribe send an email to tor-relays-leave@lists.torproject.org