Separate device log recording from the rest of the app. This should cause less strain on the DB server especially since the full event log is rarely needed; you could just use your favorite log collector to watch stderr. DB calls are still mixed, but may be refactored out later.

lib/units/log/rethinkdb.js

@@ -0,0 +1,52 @@
+var Promise = require('bluebird')
+var zmq = require('zmq')
+
+var logger = require('../../util/logger')
+var wire = require('../../wire')
+var wirerouter = require('../../wire/router')
+var wireutil = require('../../wire/util')
+var lifecycle = require('../../util/lifecycle')
+var srv = require('../../util/srv')
+var dbapi = require('../../db/api')
+
+module.exports = function(options) {
+  var log = logger.createLogger('log-db')
+
+  // Input
+  var sub = zmq.socket('sub')
+  Promise.map(options.endpoints.sub, function(endpoint) {
+    return srv.resolve(endpoint).then(function(records) {
+      return srv.attempt(records, function(record) {
+        log.info('Receiving input from "%s"', record.url)
+        sub.connect(record.url)
+        return Promise.resolve(true)
+      })
+    })
+  })
+
+  // Establish always-on channels
+  ;[wireutil.global].forEach(function(channel) {
+    log.info('Subscribing to permanent channel "%s"', channel)
+    sub.subscribe(channel)
+  })
+
+  sub.on('message', wirerouter()
+    .on(wire.DeviceLogMessage, function(channel, message) {
+      if (message.priority >= options.priority) {
+        dbapi.saveDeviceLog(message.serial, message)
+      }
+    })
+    .handler())
+
+  log.info('Listening for %s (or higher) level log messages',
+    logger.LevelLabel[options.priority])
+
+  lifecycle.observe(function() {
+    try {
+      sub.close()
+    }
+    catch (err) {
+      // No-op
+    }
+  })
+}