aboutsummaryrefslogtreecommitdiffstats
path: root/src/swarm/peer
Commit message (Collapse)AuthorAgeFilesLines
* lanes commenttv3 days1-2/+5
|
* round trips in treestv4 days1-1/+8
|
* suspect pathstv4 days1-0/+5
|
* deflate blob chunkstv4 days1-1/+2
|
* epoch tag in synctv4 days1-3/+21
|
* time blob messagestv4 days1-0/+19
|
* slow round back offtv4 days1-0/+5
|
* hello again when heard fromtv4 days2-2/+19
| | | | | | While we wait for a peer's HELLO_REPLY, anything else that comes from it makes our HELLO go again at once, at most once a second, instead of when the backoff says.
* unanswered pull frees the slottv5 days1-0/+1
| | | | | | When the answer to a pull is overdue (the sync reply timeout: 1 s in the simtests, 10 s in production), the pull gives up the slot, and the next waiting neighbour is pulled at once.
* overdue sync asked after from thentv5 days1-3/+3
|
* pull after gracetv5 days1-1/+1
| | | | | | | | Before pulling, a node waits one measured round trip to the announcer (1 s at most) and pulls only what is still missing. If the round trip isn't known yet, it pulls at once, as now. On the clean benchmark links the round trip is a few milliseconds, so those scenarios shouldn't change.
* announcements answered when lackingtv5 days2-5/+23
| | | | | | | A neighbour now answers an announcement only if it lacks some of the announced heads. Heads that merely differ are no longer enough, since they do while writes go on. The check reads the document, so it runs off the message consumer, as the pull check already does.
* unanswered members not reachedtv6 days1-7/+18
|
* replace eager links by treestv6 days1-8/+3
|
* measure 2-hop neighbourhoodtv6 days2-3/+11
|
* introduce eager linkstv6 days1-3/+8
|
* skip neighbours with our headstv6 days1-0/+1
|
* announcements answered only if heads differtv6 days3-2/+24
|
* retry new peers before probingtv6 days1-1/+1
|
* probe list memberstv6 days2-6/+7
|
* add per-node traffic countstv6 days1-1/+9
|
* probe when reachabletv7 days2-1/+26
|
* probe wakes peer workertv7 days1-2/+3
|
* peer worker forget records on changetv7 days3-3/+17
|
* peer worker sleeps until deadline, not fixed timetv7 days4-14/+80
|
* state: drop epoch_txtv7 days1-1/+2
| | | | The stamp is the only lock-free epoch label.
* rename AutoMerge to Documenttv8 days1-2/+2
|
* add peer clock skew warningtv9 days1-0/+1
|
* pong carries timetv9 days2-4/+7
|
* dashboard: try to get links dormant immediatelytv9 days1-6/+12
|
* centralize call to topology_changetv9 days4-27/+23
|
* unify request_sync logictv9 days4-16/+4
|
* unify heard_from logictv9 days2-23/+35
|
* unify closing logictv9 days1-3/+3
|
* unify sync_on predicatetv9 days1-7/+2
|
* farewell means dormanttv9 days1-1/+9
|
* add links messagestv9 days1-0/+6
| | | | | | | | | Introduce two new wire messages, LINKS and LINKS_REPLY (swarm::links). A member answers with its links: for each peer, the transport, whether it’s relayed, the RTT, and whether it’s dormant. It answers members only, at most every 5 s. This bumps wire::VERSION to 7. Every build answers, with or without the feature, so every node needs this patch before the map shows anything beyond its own node.
* immediately probe dormant members found in hellostv9 days1-4/+37
| | | | | | Instead of waiting for its regular timer, reprobe dormant members found in hello reply messages on the next worker tick. With this a partition gets reconnected quickly as soon as the first pair reconnects.
* log messages in paths layertv11 days1-2/+0
|
* topology: introduce outbound memberstv11 days1-1/+6
|
* track unlisted memberstv11 days1-0/+2
|
* peers from member list start as unknowntv12 days2-4/+4
|
* get swarm RTT from pathstv12 days1-6/+9
|
* keepalive takes turnstv12 days2-1/+17
|
* move keepalive and firewall to pathstv12 days4-30/+119
|
* measure pathstv12 days1-1/+4
|
* rename PING/PONG to STATUS/STATUS_REPLYtv12 days4-39/+41
|
* admit nodeId and addresses in the APItv13 days1-1/+8
|
* add identity layertv13 days1-2/+6
|
* introduce addressestv13 days1-3/+25
|