This should resolve the spurious ping timeouts we've had on low powered
boxes. Those errors are the result of us requiring a timely Pong
response to our Pings. However this is unnecessarily strict - as long as
we've received *anything* recently, we know the other peer is alive. So
the new mechanism removes the Pong message entirely and separates the
ping check into two routines:
- One that makes sure to send ping periodically, if nothing else has
been sent. This guarantees a message sent every 45-90 seconds.
- One that checks how long it was since we last received a message. If
it's longer than 300 seconds, we trigger an ErrTimeout.
So we're guaranteed to detect a connection failure in 300 + 300/2
seconds (due to how often the check runs) and we may detect it much
sooner if we get an actual error on the ping write (a connection reset
or so).
This is more sluggish than before but I think that's an OK price to pay
for making it actually work out of the box.
This removes the configurability of it, as the timeout on one side is
dependent on the send interval on the other side. Do we still need it
configurable?
This makes it so we can initialize the relay management and then give
that to the connection management, instead of the other way around.
This is important to me in the discovery revamp I'm doing, as otherwise
I get a circular dependency when constructing stuff, with relaying
depending on connection, connection depending on discovery, and
discovery depending on relaying.
With this fixed, discovery will depend on relaying, and connection will
depend on both discovery and relaying.
Instead, make sure we do the check as part of CheckFolderHealth before
pulling, and individually per file to try to not run out of space at
that stage.
(The latter is far from fool proof as we may pull lots of stuff in
parallell, but it's worth a try.)
This makes it possible to run multiple instances on the same box, all
receiving local discovery packets. Tested on Mac, Windows, supposed to
work on at least Linux too. For Windows, there may be issues with XP and
earlier, but meh...
Either Angular or the browser sometimes returns cached repsonse header,
causing a flap between requests that return the new version and requests
that return the old one. Here, instead, we trust the actual data
returned by the uncached /rest/system/version call.
Resetting the timeout doesn't fully cut it, as it may timeout after we
got an event and be delivered later. This should fix it well enough for
the moment. https://github.com/golang/go/issues/11513
* syncthing/pr/1995:
Add switch to disable relays
Do not start relay service unless explicitly asked for, or global announcement server is running
Add dynamic relay lookup (DDoS relays.syncthing.net!)
Discovery clients now take an announcer, global discovery is delayed
Expose connection type and relay status in the UI
Add dependencies (fixes#1364)
Check relays for available devices
Add incoming connection relay service
Add unsubscribe to config
Connections have types
Large refactoring/feature commit
I figured we're missing out on being cool and awesome by not having an
alphabetically based release code name like the big guys. This commit
fixes that. I've unilaterally decided on a theme of "$metal $bug"
because metals are kind of cool, and bugs, well, ...
The XDR unmarshaller allocates a []T when it sees a slice type and reads
the expected length, so we must always limit the length in order to
avoid allocating too much memory when encountering corruption.
We allocate a []byte to read the message into, so if the header says the
messages is several gigabytes large we may run into trouble. In reality,
a message should never be that large so we impose a limit.
This fixes a corner case I discovered in the symlink branch, where we
unexpectedly succeed in "replacing" an entire non-empty directory tree
with a file or symlink. This happens when archiving is in use, as we
then just move the entire tree away into the archive. This is wrong as
we should just archive files and fail on non-empty dirs in all cases.
New handling first checks what the (old) thing is, and if it's a
directory or symlink just does the delete, otherwise does conflict
handling or archiving as appropriate.
This will decrease the risk of running out of file descriptors for the
database and other bad things, which could otherwise potentially happen
if we're serving lots of requests and scanning in parallel, etc.
Windows doesn't have a per process open file limit like Unix so we don't
need to worry about it there.
Jakob Borg