This is the part that no document states outright, and the part that surprises

people. It follows from the pieces already quoted, but you have to put them

together yourself.

Identification is bound to a link. The manual says identify() "can only happen

once the link has been established". It says get_remote_identity() returns None

unless the initiator "has already independently called" identify(). And it says

of tearing a link down:

teardown()

Closes the link and purges encryption keys. New keys will be used if a new

link to the same destination is established.

Nothing is written to disk. There is no session, no cookie, no token, no

remembered device. The node knows who you are for exactly as long as that one

link stays up, and not one moment longer.

So your identification is lost when:

- the node restarts, which kills every link it had

- your client restarts

- the link times out from inactivity

- anything in the path drops and the link has to be rebuilt

After any of those, a new link is established and you are anonymous again until

you identify on the new link. If the page you were reading was protected by an

.allowed file, it will now refuse you, and the only clue is a terse "not

authorised" message.

Nothing is broken when this happens. It is the protocol behaving exactly as

designed. But it does mean an .allowed file is a poor lock for anything you

need to reach reliably, especially on a node you restart while working on it.

<- Prev Chapter Next Chapter ->