X-Git-Url: https://git.sesse.net/?p=remoteglot;a=blobdiff_plain;f=www%2Fserve-analysis.js;h=9bb33cfd28340d29dc7ae7e871db8b99bd148782;hp=4f0d2f010376df9748333a58a985c0fbbf1bcd21;hb=a0290e9de747b7811e8403b6e7ea80ba5d96cd71;hpb=31e240c2899bdcba193264105cc5e646565a6e73 diff --git a/www/serve-analysis.js b/www/serve-analysis.js index 4f0d2f0..9bb33cf 100644 --- a/www/serve-analysis.js +++ b/www/serve-analysis.js @@ -8,18 +8,44 @@ var url = require('url'); var querystring = require('querystring'); var path = require('path'); var zlib = require('zlib'); +var delta = require('./js/json_delta.js'); // Constants. +var HISTORY_TO_KEEP = 5; +var MINIMUM_VERSION = null; + +// Filename to serve. var json_filename = '/srv/analysis.sesse.net/www/analysis.json'; +if (process.argv.length >= 3) { + json_filename = process.argv[2]; +} + +// Expected destination filename. +var serve_url = '/analysis.pl'; +if (process.argv.length >= 4) { + serve_url = process.argv[3]; +} + +// TCP port to listen on. +var port = 5000; +if (process.argv.length >= 5) { + port = parseInt(process.argv[4]); +} + +// If set to 1, we are already processing a JSON update and should not +// start a new one. If set to 2, we are _also_ having one in the queue. +var json_lock = 0; // The current contents of the file to hand out, and its last modified time. -var json_contents = undefined; -var json_contents_gz = undefined; -var json_last_modified = undefined; +var json = undefined; -// The list of clients that are waiting for new data to show up, -// and their associated timers. Uniquely keyed by request_id -// so that we can take them out of the queue if they time out. +// The last five timestamps, and diffs from them to the latest version. +var historic_json = []; +var diff_json = {}; + +// The list of clients that are waiting for new data to show up. +// Uniquely keyed by request_id so that we can take them out of +// the queue if they close the socket. var sleeping_clients = {}; var request_id = 0; @@ -27,10 +53,88 @@ var request_id = 0; // Used to show a viewer count to the user. var last_seen_clients = {}; +// The timer used to touch the file every 30 seconds if nobody +// else does it for us. This makes sure we don't have clients +// hanging indefinitely (which might have them return errors). +var touch_timer = undefined; + +// If we are behind Varnish, we can't count the number of clients +// ourselves, so some external log-tailing daemon needs to tell us. +var viewer_count_override = undefined; + +var replace_json = function(new_json_contents, mtime) { + // Generate the list of diffs from the last five versions. + if (json !== undefined) { + // If two versions have the same mtime, clients could have either. + // Note the fact, so that we never insert it. + if (json.last_modified == mtime) { + json.invalid_base = true; + } + if (!json.invalid_base) { + historic_json.push(json); + if (historic_json.length > HISTORY_TO_KEEP) { + historic_json.shift(); + } + } + } + + var new_json = { + parsed: JSON.parse(new_json_contents), + plain: new_json_contents, + last_modified: mtime + }; + create_json_historic_diff(new_json, historic_json.slice(0), {}, function(new_diff_json) { + // gzip the new version (non-delta), and put it into place. + zlib.gzip(new_json_contents, function(err, buffer) { + if (err) throw err; + + new_json.gzip = buffer; + json = new_json; + diff_json = new_diff_json; + json_lock = 0; + + // Finally, wake up any sleeping clients. + possibly_wakeup_clients(); + }); + }); +} + +var create_json_historic_diff = function(new_json, history_left, new_diff_json, cb) { + if (history_left.length == 0) { + cb(new_diff_json); + return; + } + + var histobj = history_left.shift(); + var diff = delta.JSON_delta.diff(histobj.parsed, new_json.parsed); + var diff_text = JSON.stringify(diff); + zlib.gzip(diff_text, function(err, buffer) { + if (err) throw err; + new_diff_json[histobj.last_modified] = { + parsed: diff, + plain: diff_text, + gzip: buffer, + last_modified: new_json.last_modified, + }; + create_json_historic_diff(new_json, history_left, new_diff_json, cb); + }); +} + var reread_file = function(event, filename) { if (filename != path.basename(json_filename)) { return; } + if (json_lock >= 2) { + return; + } + if (json_lock == 1) { + // Already processing; wait a bit. + json_lock = 2; + setTimeout(function() { json_lock = 1; reread_file(event, filename); }, 100); + return; + } + json_lock = 1; + console.log("Rereading " + json_filename); fs.open(json_filename, 'r+', function(err, fd) { if (err) throw err; @@ -41,24 +145,27 @@ var reread_file = function(event, filename) { if (err) throw err; fs.close(fd, function() { var new_json_contents = buffer.toString('utf8', 0, bytesRead); - zlib.gzip(new_json_contents, function(err, buffer) { - if (err) throw err; - json_contents = new_json_contents; - json_contents_gz = buffer; - json_last_modified = st.mtime.getTime(); - possibly_wakeup_clients(); - }); + replace_json(new_json_contents, st.mtime.getTime()); }); }); }); }); + + if (touch_timer !== undefined) { + clearTimeout(touch_timer); + } + touch_timer = setTimeout(function() { + console.log("Touching analysis.json due to no other activity"); + var now = Date.now() / 1000; + fs.utimes(json_filename, now, now); + }, 30000); } var possibly_wakeup_clients = function() { var num_viewers = count_viewers(); for (var i in sleeping_clients) { - clearTimeout(sleeping_clients[i].timer); mark_recently_seen(sleeping_clients[i].unique); send_json(sleeping_clients[i].response, + sleeping_clients[i].ims, sleeping_clients[i].accept_gzip, num_viewers); } @@ -71,37 +178,58 @@ var send_404 = function(response) { response.write('Something went wrong. Sorry.'); response.end(); } -var send_json = function(response, accept_gzip, num_viewers) { +var handle_viewer_override = function(request, u, response) { + // Only accept requests from localhost. + var peer = request.socket.localAddress; + if ((peer != '127.0.0.1' && peer != '::1') || request.headers['x-forwarded-for']) { + console.log("Refusing viewer override from " + peer); + send_404(response); + } else { + viewer_count_override = (u.query)['num']; + response.writeHead(200, { + 'Content-Type': 'text/plain', + }); + response.write('OK.'); + response.end(); + } +} +var send_json = function(response, ims, accept_gzip, num_viewers) { + var this_json = diff_json[ims] || json; + var headers = { 'Content-Type': 'text/json', - 'X-Remoteglot-Last-Modified': json_last_modified, - 'X-Remoteglot-Num-Viewers': num_viewers, - 'Access-Control-Allow-Origin': 'http://analysis.sesse.net', - 'Access-Control-Expose-Headers': 'X-Remoteglot-Last-Modified, X-Remoteglot-Num-Viewers', - 'Expires': 'Mon, 01 Jan 1970 00:00:00 UTC', + 'X-RGLM': this_json.last_modified, + 'X-RGNV': num_viewers, + 'Access-Control-Expose-Headers': 'X-RGLM, X-RGNV, X-RGMV', + 'Vary': 'Accept-Encoding', }; + if (MINIMUM_VERSION) { + headers['X-RGMV'] = MINIMUM_VERSION; + } + if (accept_gzip) { + headers['Content-Length'] = this_json.gzip.length; headers['Content-Encoding'] = 'gzip'; response.writeHead(200, headers); - response.write(json_contents_gz); + response.write(this_json.gzip); } else { + headers['Content-Length'] = this_json.plain.length; response.writeHead(200, headers); - response.write(json_contents); + response.write(this_json.plain); } response.end(); } -var timeout_client = function(client) { - mark_recently_seen(client.unique); - send_json(client.response, client.accept_gzip, count_viewers()); - delete sleeping_clients[client.request_id]; -} var mark_recently_seen = function(unique) { if (unique) { last_seen_clients[unique] = (new Date).getTime(); } } var count_viewers = function() { + if (viewer_count_override !== undefined) { + return viewer_count_override; + } + var now = (new Date).getTime(); // Go through and remove old viewers, and count them at the same time. @@ -137,9 +265,12 @@ server.on('request', function(request, response) { var ims = (u.query)['ims']; var unique = (u.query)['unique']; - console.log((new Date).getTime()*1e-3 + " " + request.url); - - if (u.pathname !== '/analysis.pl') { + console.log(((new Date).getTime()*1e-3).toFixed(3) + " " + request.url); + if (u.pathname === '/override-num-viewers') { + handle_viewer_override(request, u, response); + return; + } + if (u.pathname !== serve_url) { // This is not the request you are looking for. send_404(response); return; @@ -157,20 +288,19 @@ server.on('request', function(request, response) { // If we already have something newer than what the user has, // just send it out and be done with it. - if (json_last_modified !== undefined && (!ims || json_last_modified > ims)) { - send_json(response, accept_gzip, count_viewers()); + if (json !== undefined && (!ims || json.last_modified > ims)) { + send_json(response, ims, accept_gzip, count_viewers()); return; } // OK, so we need to hang until we have something newer. - // Put the user on the wait list; if we don't get anything - // in 30 seconds, though, we'll send something anyway. + // Put the user on the wait list. var client = {}; client.response = response; - client.timer = setTimeout(function() { timeout_client(client); }, 30000); client.request_id = request_id; client.accept_gzip = accept_gzip; client.unique = unique; + client.ims = ims; sleeping_clients[request_id++] = client; request.socket.client = client; @@ -184,4 +314,5 @@ server.on('connection', function(socket) { } }); }); -server.listen(5000); + +server.listen(port);