reweave/weave
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
1,403 Commits 7 Branches 85 Tags
8a2828d57d4938d7a9852cb33d0cd6fb433ffd22
Commit Graph
251 Commits
Author SHA1 Message Date
Matthias Radestock 8a2828d57d add "representation of received update" to return of Peers.ApplyUpdate 2015-04-06 14:17:23 +01:00
Matthias Radestock a58eb6ba7d make gossip broadcast operate on GossipData 2015-04-06 14:00:51 +01:00
Matthias Radestock 9b9d8a03aa refactor: make Peer.Connections() return a set instead of a slice 
This is more flexible since it allows callers to perform efficient
lookups on the result.
 2015-04-06 13:57:13 +01:00
Matthias Radestock e8e9a43abd tired of typing "struct{}{}" 2015-04-06 11:53:51 +01:00
Matthias Radestock 6db12d682e make PeerNameSet a more broadly used type 2015-04-06 11:45:17 +01:00
Matthias Radestock 74b3f8d44e use map[...]struct{} for sets instead of map[...]bool 
See https://groups.google.com/d/msg/golang-nuts/lb4xLHq7wug/MhrSLkyS4F8J

We don't really care about the performance aspect of this, but using
struct{}{} values is in some way canonical since it means the
representation carries no redundant information.
 2015-04-06 11:35:11 +01:00
Matthias Radestock 624373ace2 simplify 2015-04-05 17:03:18 +01:00
Matthias Radestock 8c1015d2a4 cosmetic: consistent terminology in var names 2015-04-04 22:18:31 +01:00
Matthias Radestock 95c1e226f0 use routes based on tcp overlay connectivity for gossip 
i.e. include asymmetric and non-established connections
This is TRTTD since gossip flows over the tcp overlay.
 2015-04-04 21:56:30 +01:00
Matthias Radestock 3fba123bc0 comment 2015-04-04 21:14:25 +01:00
Matthias Radestock b272dcf8c1 rename: less 'gossip' in gossip 2015-04-04 20:58:10 +01:00
Matthias Radestock 91ed22a953 refactor(ish): generalise an abstraction 
LocalPeer.NextBroadcastHops -> ConnectionsTo
thus separating the computation of who we want to connect to from
obtaining the associated Connections
 2015-04-04 20:36:48 +01:00
Matthias Radestock bcd2a0184c small refactor 2015-04-04 20:08:38 +01:00
Matthias Radestock 45fd750318 cosmetic: move some test helpers from test code into main code 
since they reach quite deeply into the data structures
 2015-04-04 20:02:22 +01:00
Matthias Radestock 0a02d1a926 cosmetic: rename a type 2015-04-04 20:02:01 +01:00
Matthias Radestock ac0788cb5a simplify test router creation 2015-04-04 19:50:19 +01:00
Matthias Radestock 683caef866 tweak comment 2015-04-04 19:46:47 +01:00
Matthias Radestock fa5279ab3e cosmetic 2015-04-04 16:02:01 +01:00
Matthias Radestock 6cda74d49f refactor: inline 2015-04-04 15:36:42 +01:00
Matthias Radestock fdea7d93f9 refactor: improve abstraction in connection gossip handling 
- turn previously free-floating functions into methods on GossipChannel
- move gossip message type dispatch from connection into router
 2015-04-03 19:38:01 +01:00
Matthias Radestock 6f398a7459 refactor(ish): raise GossipSender abstraction level slightly 2015-04-03 18:33:59 +01:00
Matthias Radestock 7166d151aa make route calculcation more efficient and properly async 
Instead of following the usual actor pattern, we use a 1-place channel
with a non-blocking send. Essentially, if there is already a pending
recalculation request queued up, there is no point enqueuing another
one.
 2015-04-03 18:12:03 +01:00
David Wragg 587a67d701 Comment 
It's not blindingly obvious where this 'found' comes from.
 2015-04-02 16:37:30 +01:00
Matthias Radestock 5648f87717 oops 
broke this in 45740b1419
 2015-04-02 13:03:42 +01:00
Matthias Radestock e97a3f88aa allow router to be started w/o capture/inject interface 
useful for testing of the control plane
 2015-04-02 09:46:29 +01:00
Matthias Radestock e33d2eaf8e cosmetic 2015-04-02 07:10:21 +01:00
Matthias Radestock b50688b4de act immediately on weave connecting existing addresses 
we simply reset the vars controlling backoff for that address
Note that we do not interrupt any current connection attempt, since
that is hard.

Closes #499.
 2015-03-31 11:05:11 +01:00
David Wragg b00be096f7 Fix returns when iterating over Connections() 
d5eb76f32 replaced (*Peer)ForEachConnection into (*Peer)Connections(),
and consequently converted some nested functions into for loops.  But
the "return" statements in those nested functions were left as-was,
and so exited the outer function rather than moving on to the next
iteration.

Fixes #497
 2015-03-29 19:10:44 +01:00
Matthias Radestock d708cb7d4e prevent PMTUVerfied handling from blocking 
...due to blocking on sending the pmtu to the forwarder, because the
forwarder has terminated.

Fixes #464.
 2015-03-29 14:04:28 +01:00
Matthias Radestock fc22f9664e refactor: improve encapsulation of Forwarder 2015-03-29 12:14:32 +01:00
Matthias Radestock 07c11d87a8 oops 
I broke this in 1eca5e1097
 2015-03-29 12:11:59 +01:00
Matthias Radestock 4eda7e361b put back forwarder channel nil check 
since otherwise forward(...) can block indefinitely due to both the
forwarder channel and finshed channel being nil
 2015-03-29 10:36:01 +01:00
Matthias Radestock b6448fa196 more idiomatic closing of forwarders, by closing a 'finished' channel 
The previous scheme set the reference to main channel to nil and then
drained it. That was racy and hence could lead to deadlock in the
following scenario:

1. multiple senders obtain the main channel from the stored reference
2. main channel reference is set to nil
3. main channel is drained
4. multiple senders send to main channel

So the issue here is that the entire forwarder stopping sequence
occurs between senders obtaining of the reference to the main channel,
and then sending on that channel. If more senders sent than there is
capacity in the channel then this leads to deadlock.
 2015-03-28 20:58:42 +00:00
Matthias Radestock 4dbb3d97e0 cosmetic: remove dead code 2015-03-28 20:33:06 +00:00
Matthias Radestock 502b59c703 refactor: rename 2015-03-28 19:52:10 +00:00
Matthias Radestock 9a462f9d87 refactor: simplify NewNaClEncryptor 
- rely on default nil values on initialisation
- inline
 2015-03-28 19:49:01 +00:00
Matthias Radestock 9992992ee4 refactor: extract helper from NaClDecryptor.decrypt 
for easier comprehension
 2015-03-28 19:44:04 +00:00
David Wragg 78c70023d8 Don't pass the sender UDPAddr through the crypto layer 
The crypto layer doesn't need it, and the cost of an additional
closure allocation for each incoming UDP packet is modest.

Also, make the FrameConsumer the last parameter to IterateFrames.
It seems a more natural ordering.
 2015-03-27 18:05:00 +00:00
David Wragg 0d30df165f Eliminate UDPPacket 
The Name field is not used anywhere, and once that's gone the remaining
pairing of the packet bytes and sender is not worth it.
 2015-03-27 18:05:00 +00:00
David Wragg 4a1d3aeb96 Remove frameLen parameter to FrameConsumer 
It's the same as len(frame)
 2015-03-27 18:05:00 +00:00
David Wragg bc45a363dd Make Encryptor.AppendFrame more consistent with FrameConsumer 
Also removes references to Peer type in crypto.go, to keep it agnostic
of higher-level concerns.
 2015-03-27 18:05:00 +00:00
David Wragg a670272a37 Remove name parameter from (*Peers)ForEach consumer function 
It's part of the Peer anyway.
 2015-03-27 18:05:00 +00:00
Matthias Radestock a337e981ba refactor: extract helper 2015-03-27 12:38:06 +00:00
Matthias Radestock 0ef717a7a2 refactor: extract helper function 2015-03-27 12:31:25 +00:00
Matthias Radestock d25d295784 refactor: small tweak for consistency 2015-03-27 12:17:28 +00:00
Matthias Radestock 60ba8cb90d refactor: extract helper function 2015-03-27 12:06:22 +00:00
Matthias Radestock 4d8c156915 continuing accumulation after dropping a frame is bad after all 
explain why
 2015-03-27 11:56:38 +00:00
Matthias Radestock 6ea625f85b fix typo 2015-03-27 11:54:24 +00:00
Matthias Radestock 84a95f2daf refactor: concentrate frame appending code in one place 2015-03-27 11:53:56 +00:00
Matthias Radestock f544fc68de enter frame accumulation even after dropping one 
This logically makes more sense, i.e. we keep draining frames from
fwd.ch until we run out.
 2015-03-27 11:35:42 +00:00