mirror of
https://github.com/amark/gun.git
synced 2025-03-30 15:08:33 +00:00

* tmp for hn issue * log top to stats * test for guntest peer * try big messages * parse time? * what bin/node is 11ms? * be normal for hnoon * tolerate 0.5s * try 3s for hnoon? * stop empty gets * tmp for guntest * back to normal * check hash time * back to normal in hear * screen / upload / play / pause * merge latest npm release into manhattan * merge master 0.2020.421 into manhattan manually * WIP * manually merge from master * gatling * Update upload.html * work in progress... * yson panic chat basic * after `.put(` walk * restructure acks * messy but 2 units passing! * put recursive once on map * basics * have NTS use DAM + fix other utils * Rewrote nts * Allow passing test cli args. Before, no CLI args would be passed when running `npm test`. Keeping the `mocha` at the end of the test script allows passing CLI args to Mocha. * put back scan & once tweak * PANIC user paste OK * manhattan sea * stub out nts for now * AXE tweak * tweak for quick first prod testing * tweak for first in-prod testing * tweak * tweak * sketchy in-prod debug attempt * caught it? maybe? now restore * Create download-log.html * stub out yson test prod? * ugh, gotta see what is going on * move dl * gonna stop doing commit messages for in-prod test/tweaks/debugging * a * p * squelch * console stats * stats * stop travis * restore yson * ahhh no file access without sudo * mem * no stub * fix axe * bump * back to in-prod testing, isolate/stub out code * stub all out for 17K ? CPU ? test * stub dup gc * ugh main stub * does this stop url format blocking? * re-add dup * no top :( * will this work? * get ack stats? * a map chain may ask for data not a root soul chain * move proper logic into .get( * how 2019 compat? * a couple more! * more tests passing! :D :) * even more! SO EXCITING :D * Am I alive? * wow I can't believe it works like this * THANK YOU @rogowski !!!!!! * Create trace.html Adding tracing to debuging. * @rogowski is a super star :) :) :) * Update trace.js Change `Gun.logs` to `Gun.traces` and `Gun._log` to `Gun._trace`. * Update trace.html Change `Gun.logs` to `Gun.traces` and `Gun._log` to `Gun._trace`. Overload get,put,on,map * @rogowski approved of these trace changes :) * Update trace.html More decoupled. * Update trace.js More decoupled * 2 steps backwards, 1 step forward? * back where we ( @rogowski ) started :P * YAYAYAYAYAYAYAY past where we started at! * safer to have it here * slight tweak? Let's see how long it lasts. * merge checks we left out during consolidation * ugly common.js for @rogowski * slightly better * amazing map discovery + don't clear on not found if data exists * onto next test... * all caught up!!! Now update tests from graphify to statedisk * Update common.js Tests updated from graphify to statedisk. * easy to debug & fix thanks to @rogowski 's test upgrades & trace! * hmm, IDK if we should support this anymore? * support once chaining? * check if listener order is earlier than write * in-process message passing needs to clean itself of flags for now * ack to chains that can't be reached * call sub chains on clear/empty WIP * call sub chains clear/empty OK! * into unlink. Clean/refactor later. * oh that was nice * self check not needed? * test was poorly constructed? * refactor unlink to cleaner logic * Will you blame me for this? Special case, maybe later move to cleaner place? * use stun's run id instead. * cleaner unlink logic * better map, link, and unlink logic. * unstub relay * refactor unlink * invert * if prev value would have caused unlink, do not unlink again. * w000h00! Best unlink so far. * woops, fix unlinking nested * unsubscribe nested maps (working, tho possible perf regression? check) * put check soul * add default uuid * improved browser peer retry logic, let devices sleep, etc. * Chaining API unit tests passing! * merge new panic tests into here to test * add panic utils * fix long streaming timeout/expiry issue, update examples * yield generating test data * yeah, adapter errors (like out of storage) should not affect sync/networking logic, that was a bad experiment * git glitch? * some mid debugging fixes but maybe scary changes, hopefully safe to revert here except dub * SEA unit tests passing!!! Needed quite a few fixes on async write queue stuff. * optionally make auth async * revise/correct set * Fix reverse boundary check * Add extra tests, catch bad guy, obliterate bug. * chat app with emoji examples * handle empty string keyed objects * starting lex support * tweak for lex * woops! lexical alphabetical oopsies. That was bad. * upload either way * debug * start * fix * fix * clean + feature * update dependencies in package.json (#1086) * rad lex once map once * axe polyfill for now * oops log * oops maybe without this it crashed the peer * what on earth happened to my browser/OS? "unplug & plug it back in" restart seemed to fix it. * oh, don't memory leak req/res asks. :/ duh! * no accidental #soul."" * ugh, still have to sort :(, really should polyfill weakmap then * oops, pluck needs new object to go into * oops, make sure soul is passed * updating deprecated functions * begin AXE. Next: load balance! * Update sea.js * keys are dangerous! * AXE round robin load balance * better ash hash checking * lS reuse in-mem reply chunking * state machine!!! * RAD needs to pass cache misses. * updating deprecated functions (#1088) * update dependencies in package.json * updating deprecated functions * remove where.gundb.io * Bring SEA.certify into manhattan branch (#1092) Co-authored-by: Radu Cioienaru <radu@projectmanager.com> * fix rad, make get() hookable * rad browser tests seem to be passing! * reverse user random side, add err, update styles, + more * fix pack/max, update dom * paste! * of course it'll dedup cause it just called track on hear, fix * 📦 Adding the hub feature to this branch & improvements. (#1102) * 📦 Adding the hub feature to this branch. * 🗑 Removed the container for speed improvement ! * 📝 I added some comments to the code. Co-authored-by: Hector <fairfairytotor@gmail.com> Co-authored-by: Hector <pro.hector.kub@gmail.com> * Update axe.js * 🦅 Wrap everything in a try & catch for error handling… (#1105) * 🦅 Wrap everything in a try & catch for error handling & speed improvement. * 📦 Finally here : opt.file for the hub feature ! * 📦 Finally here : opt.file for the hub feature ! And also : fixed indentation 😋 Co-authored-by: noctisatrae <pro.hector.kub@gmail.com> * probs better this way, safer * moved test/axe tests to test/panic/axe. * New test: axe load balance. * axe test: webrtc data balance(fix paths and file renamed). * test axe: renaming webrtc file. * axe test: separating webrtc test for data_balance. * axe test: test only with the relay(without webrtc). * Update sea.js Same as https://github.com/amark/gun/pull/1062 * Update gun.js var tmp * Update upload.js * merge, update stun * SEA.certify wire logic + unit tests (#1110) * SEA.certify wire logic + unit tests * picking white hair * ack err * axe tests using puppeteer. * change stun system * ~20lines * put use parent soul link if need * handle errors * finally seems fixed * cb not to * relay * nasty bug! Don't crash, tho need to find what causes it * undo local changes/notes to self * deprecation warnings * "old" data to test against * oops, forgot I played with ascii * debug * in-prod check: sites * in-prod isolate * gotta find this, by stubbing out * where? * will this work? * clearly not, lol what's the point then? maybe like this * and again * must we? * USE THIS MANHATTAN VERSION * clean * better panic hints Co-authored-by: Robin Bron <finwo@pm.me> Co-authored-by: Pavel Diatchenko <diatche@users.noreply.github.com> Co-authored-by: rogowski <163828+rogowski@users.noreply.github.com> Co-authored-by: I001962 <i001962@gmail.com> Co-authored-by: Adriano Rogowski <rogowski.adriano@gmail.com> Co-authored-by: Radu <cetatuie@gmail.com> Co-authored-by: Radu Cioienaru <radu@projectmanager.com> Co-authored-by: Hector <46224745+noctisatrae@users.noreply.github.com> Co-authored-by: Hector <fairfairytotor@gmail.com> Co-authored-by: Hector <pro.hector.kub@gmail.com> Co-authored-by: Martti Malmi <sirius@iki.fi> Co-authored-by: mimiza <dev@mimiza.com>
84 lines
3.0 KiB
JavaScript
84 lines
3.0 KiB
JavaScript
var Gun = require('../gun');
|
|
|
|
/*
|
|
An Ad-Hoc Mesh-Network Daisy-Chain
|
|
should work even if humans are
|
|
communicating with each other blind.
|
|
|
|
To prevent infinite broadcast loops,
|
|
we use a deduplication process
|
|
based on the message's identifier.
|
|
This is currently implemented in core.
|
|
|
|
However, because this still creates a
|
|
N*2 (where N is the number of connections)
|
|
flood, it is not scalable for traditional
|
|
services that have a hub network topology.
|
|
|
|
Does this mean we have to abandon mesh
|
|
algorithms? No, we can simply layer more
|
|
efficient optimizations in based on constraints.
|
|
If these constraints exist, it automatically
|
|
upgrades, but if not, it falls back to the
|
|
brute-force mesh based robust algorithm.
|
|
A simple example is to limit peer connections
|
|
and rely upon daisy chaining to relay messages.
|
|
|
|
Another example, is if peers are willing to
|
|
identify themselves, then we can improve the
|
|
efficiency of the network by having each peer
|
|
include the names of peers it is connected in
|
|
each message. Then each subsequent peer will
|
|
not relay it to them, since it is unnecessary.
|
|
This should create N (where N is the number of
|
|
peers) messages (or possibly N+ if there is a
|
|
common peer of uncommon peers that receives it
|
|
and relays at exact latency timings), which is
|
|
optimal.
|
|
|
|
Since computer networks aren't actually blind,
|
|
we will implement the above method to improve
|
|
the performance of the ad-hoc mesh network.
|
|
|
|
But why not have every message contain the
|
|
whole history of peers that it relayed through?
|
|
Because in sufficiently large enough networks,
|
|
with extensive daisy chaining, this will cause
|
|
the message to become prohibitively slow and
|
|
increase indefinitely in size.
|
|
|
|
*/
|
|
|
|
Gun.on('opt', function(root){
|
|
var opt = root.opt;
|
|
if(false === opt.ws || opt.once){
|
|
this.to.next(root);
|
|
return;
|
|
}
|
|
|
|
var url = require('url');
|
|
opt.mesh = opt.mesh || Gun.Mesh(root);
|
|
opt.WebSocket = opt.WebSocket || require('ws');
|
|
var ws = opt.ws = opt.ws || {};
|
|
ws.path = ws.path || '/gun';
|
|
// if we DO need an HTTP server, then choose ws specific one or GUN default one.
|
|
if(!ws.noServer){
|
|
ws.server = ws.server || opt.web;
|
|
if(!ws.server){ this.to.next(root); return } // ugh, bug fix for @jamierez & unstoppable ryan.
|
|
}
|
|
ws.web = ws.web || new opt.WebSocket.Server(ws); // we still need a WS server.
|
|
ws.web.on('connection', function(wire, req){ var peer;
|
|
wire.headers = wire.headers || (req||'').headers || '';
|
|
console.STAT && ((console.STAT.sites || (console.STAT.sites = {}))[wire.headers.origin] = 1);
|
|
opt.mesh.hi(peer = {wire: wire});
|
|
wire.on('message', function(msg){
|
|
opt.mesh.hear(msg.data || msg, peer);
|
|
});
|
|
wire.on('close', function(){
|
|
opt.mesh.bye(peer);
|
|
});
|
|
wire.on('error', function(e){});
|
|
setTimeout(function heart(){ if(!opt.peers[peer.id]){ return } try{ wire.send("[]") }catch(e){} ;setTimeout(heart, 1000 * 20) }, 1000 * 20); // Some systems, like Heroku, require heartbeats to not time out. // TODO: Make this configurable? // TODO: PERF: Find better approach than try/timeouts?
|
|
});
|
|
this.to.next(root);
|
|
}); |