close streams whenever we finish using them. time out if necessary

lib/storage/README.md

@@ -1,59 +0,0 @@
-# Storage Mechanisms
-
-Cryptpad's message API is quite simple and modular, and it isn't especially difficult to write alternative modules that employ your favourite datastore.
-
-There are a few guidelines for creating a module:
-
-Dependencies for your storage engine **should not** be added to Cryptpad.
-Instead, write an adaptor, and place it in `cryptpad/storage/yourAdaptor.js`.
-
-Alternatively, storage adaptors can be published to npm, and required from your config (once installed).
-
-## Your adaptor should conform to a simple API.
-
-It must export an object with a single property, `create`, which is a function.
-That function must accept two arguments:
-
-1. an object containing configuration values
-  - any configuration values that you require should be well documented
-  - they should also be named carefully so as to avoid collisions with other modules
-2. a callback
-  - this callback is used to return an object with (currently) two methods
-  - even if your storage mechanism can be executed synchronously, we use the callback pattern for portability.
-
-## Methods
-
-### message(channelName, content, handler)
-
-When Cryptpad receives a message, it saves it into its datastore using its equivalent of a table for its channel name, and then relays the message to every other client which is participating in the same channel.
-
-Relaying logic exists outside of the storage module, you simply need to store the message then execute the handler on success.
-
-### getMessages(channelName, handler, callback)
-
-When a new client joins, they request the entire history of messages for a particular channel.
-This method retreives those messages, and delivers them in order.
-
-In practice, out of order messages make your clientside application more likely to fail, however, they are generally tolerated.
-As a channel accumulates a greater number of messages, the likelihood of the application receiving them in the wrong order becomes greater.
-This results in older sessions becoming less reliable.
-
-This function accepts the name of the channel in which the user is interested, the handler for each message, and the callback to be executed when the last message has been fetched and handled.
-
-**Note**, the callback is a new addition to this API.
-It is only implemented within the leveldb adaptor, making our latest code incompatible with the other back ends.
-While we migrate to our new Netflux API, only the leveldb adaptor will be supported.
-
-## removeChannel(channelName, callback)
-
-This method is called (optionally, see config.example.js for more info) some amount of time after the last client in a channel disconnects.
-
-It should remove any history of that channel, and execute a callback which takes an error message as an argument.
-
-## Documenting your adaptor
-
-Naturally, you should comment your code well before making a PR.
-Failing that, you should definitely add notes to `cryptpad/config.example.js` such that people who wish to install your adaptor know how to do so.
-
-Notes on how to install the back end, as well as how to install the client for connecting to the back end (as is the case with many datastores), as well as how to configure cryptpad to use your adaptor.
-The current configuration file should serve as an example of what to add, and how to comment.

lib/storage/file.js

@@ -58,6 +58,23 @@ var channelExists = function (filepath, cb) {
     });
 };
 
+const destroyStream = function (stream) {
+    stream.close();
+    setTimeout(function () {
+        try { stream.destroy(); } catch (err) { console.log(err); }
+    }, 5000);
+};
+
+const ensureStreamCloses = function (stream, id, ms) {
+    return Util.bake(Util.mkTimeout(Util.once(function (err) {
+        destroyStream(stream, id);
+        if (err) {
+            // this can only be a timeout error...
+            console.log("stream close error:", err, id);
+        }
+    }), ms || 15000), []);
+};
+
 // readMessagesBin asynchronously iterates over the messages in a channel log
 // the handler for each message must call back to read more, which should mean
 // that this function has a lower memory profile than our classic method
@@ -65,9 +82,10 @@ var channelExists = function (filepath, cb) {
 // it also allows the handler to abort reading at any time
 const readMessagesBin = (env, id, start, msgHandler, cb) => {
     const stream = Fs.createReadStream(mkPath(env, id), { start: start });
+    const finish = ensureStreamCloses(stream, id);
     return void readFileBin(stream, msgHandler, function (err) {
-        try { stream.close(); } catch (err2) { }
         cb(err);
+        finish();
     });
 };
 
@@ -75,24 +93,13 @@ const readMessagesBin = (env, id, start, msgHandler, cb) => {
 // returns undefined if the first message was not an object (not an array)
 var getMetadataAtPath = function (Env, path, _cb) {
     const stream = Fs.createReadStream(path, { start: 0 });
-
-    // cb implicitly destroys the stream, if it exists
-    // and calls back asynchronously no more than once
-    /*
-    var cb = Util.once(Util.both(function () {
-        try {
-            stream.destroy();
-        } catch (err) {
-            return err;
-        }
-    }, Util.mkAsync(_cb)));
-    */
-
-    var cb = Util.once(Util.mkAsync(_cb), function () {
+    const finish = ensureStreamCloses(stream, path);
+    var cb = Util.once(Util.mkAsync(Util.both(_cb, finish)), function () {
         throw new Error("Multiple Callbacks");
     });
 
     var i = 0;
+
     return readFileBin(stream, function (msgObj, readMore, abort) {
         const line = msgObj.buff.toString('utf8');
 
@@ -130,7 +137,8 @@ var closeChannel = function (env, channelName, cb) {
     if (!env.channels[channelName]) { return void cb(); }
     try {
         if (typeof(Util.find(env, [ 'channels', channelName, 'writeStream', 'close'])) === 'function') {
-            env.channels[channelName].writeStream.close();
+            var stream = env.channels[channelName].writeStream;
+            destroyStream(stream, channelName);
         }
         delete env.channels[channelName];
         env.openFiles--;
@@ -172,11 +180,15 @@ var clearChannel = function (env, channelId, _cb) {
 */
 var readMessages = function (path, msgHandler, _cb) {
     var stream = Fs.createReadStream(path, { start: 0});
-    var cb = Util.once(Util.mkAsync(_cb));
+    const finish = ensureStreamCloses(stream, path);
+    var cb = Util.once(Util.mkAsync(Util.both(finish, _cb)));
+
     return readFileBin(stream, function (msgObj, readMore) {
         msgHandler(msgObj.buff.toString('utf8'));
         readMore();
-    }, cb);
+    }, function (err) {
+        cb(err);
+    });
 };
 
 /*  getChannelMetadata
@@ -192,9 +204,13 @@ var getChannelMetadata = function (Env, channelId, cb) {
 };
 
 // low level method for getting just the dedicated metadata channel
-var getDedicatedMetadata = function (env, channelId, handler, cb) {
+var getDedicatedMetadata = function (env, channelId, handler, _cb) {
     var metadataPath = mkMetadataPath(env, channelId);
     var stream = Fs.createReadStream(metadataPath, {start: 0});
+
+    const finish = ensureStreamCloses(stream, metadataPath);
+    var cb = Util.both(finish, _cb);
+
     readFileBin(stream, function (msgObj, readMore) {
         var line = msgObj.buff.toString('utf8');
         try {
@@ -679,11 +695,12 @@ export type ChainPadServer_ChannelInternal_t = {
     path: string
 };
 */
-var getChannel = function (
+var getChannel = function ( // XXX BatchRead
     env,
     id,
     _callback /*:(err:?Error, chan:?ChainPadServer_ChannelInternal_t)=>void*/
 ) {
+    //console.log("getting channel [%s]", id);
     var callback = Util.once(Util.mkAsync(_callback));
     if (env.channels[id]) {
         var chan = env.channels[id];
@@ -702,6 +719,7 @@ var getChannel = function (
         // if you're running out of open files, asynchronously clean up expired files
         // do it on a shorter timeframe, though (half of normal)
         setTimeout(function () {
+            //console.log("FLUSHING UNUSED CHANNELS");
             flushUnusedChannels(env, function () {
                 if (env.verbose) {
                     console.log("Approaching open file descriptor limit. Cleaning up");
@@ -740,7 +758,7 @@ var getChannel = function (
             fileExists = exists;
         }));
     }).nThen(function (waitFor) {
-        var stream = channel.writeStream = Fs.createWriteStream(path, { flags: 'a' });
+        var stream = channel.writeStream = Fs.createWriteStream(path, { flags: 'a' }); // XXX
         env.openFiles++;
         stream.on('open', waitFor());
         stream.on('error', function (err /*:?Error*/) {
@@ -762,12 +780,12 @@ var getChannel = function (
 // write a message to the disk as raw bytes
 const messageBin = (env, chanName, msgBin, cb) => {
     var complete = Util.once(cb);
-    getChannel(env, chanName, function (err, chan) {
+    getChannel(env, chanName, function (err, chan) { // XXX
         if (!chan) { return void complete(err); }
         chan.onError.push(complete);
         chan.writeStream.write(msgBin, function () {
             chan.onError.splice(chan.onError.indexOf(complete), 1);
-            chan.atime = +new Date();
+            chan.atime = +new Date(); // XXX we should just throttle closing, much simpler and less error prone
             complete();
         });
     });
@@ -781,34 +799,23 @@ var message = function (env, chanName, msg, cb) {
 // stream messages from a channel log
 // TODO replace getMessages with readFileBin
 var getMessages = function (env, chanName, handler, cb) {
-    getChannel(env, chanName, function (err, chan) {
-        if (!chan) {
-            cb(err);
-            return;
-        }
     var errorState = false;
-        readMessages(chan.path, function (msg) {
+    var path = mkPath(env, chanName);
+    readMessages(path, function (msg) {
         if (!msg || errorState) { return; }
-            //console.log(msg);
         try {
             handler(msg);
         } catch (e) {
             errorState = true;
-                return void cb(err);
+            return void cb(e);
         }
     }, function (err) {
         if (err) {
             errorState = true;
             return void cb(err);
         }
-            // is it really, though? what if we hit the limit of open channels
-            // and 'clean up' in the middle of reading a massive file?
-            // certainly unlikely
-            if (!chan) { throw new Error("impossible, flow checking"); }
-            chan.atime = +new Date();
         cb();
     });
-    });
 };
 
 var trimChannel = function (env, channelName, hash, _cb) {
@@ -832,12 +839,7 @@ var trimChannel = function (env, channelName, hash, _cb) {
     var ABORT;
 
     var cleanUp = function (cb) {
-        if (tempStream && !tempStream.closed) {
-            try {
-                tempStream.close();
-            } catch (err) { }
-        }
-
+        destroyStream(tempStream);
         Fse.unlink(tempChannelPath, function (err) {
             // proceed if deleted or if there was nothing to delete
             if (!err || err.code === 'ENOENT') { return cb(); }