Auto merge of #593 - Mark-Simulacrum:logging, r=Mark-Simulacrum

Avoid expensive queries at startup

Previously, the initial startup would synchronize() the configuration, and
query the experiment assigned to each agent. Querying the assigned experiment
could take upwards of ~80 seconds per agent, leading to a long time before it
was able to respond to queries.

I plan to see if we can restructure that query separately, but this seems like a good start.

Author: bors

src/server/agents.rs

@@ -86,12 +86,19 @@ impl Agents {
     fn synchronize(&self, tokens: &Tokens) -> Fallible<()> {
         self.db.transaction(|trans| {
             let mut real = tokens.agents.values().collect::<HashSet<&String>>();
-            for agent in &self.all()? {
-                if !real.remove(&agent.name) {
-                    trans.execute("DELETE FROM agents WHERE name = ?1;", &[&agent.name])?;
+            let current: Vec<String> = self
+                .db
+                .query("select name from agents;", [], |r| r.get(0))?;
+
+            // If the token is no longer configured, then drop this agent from
+            // our list.
+            for current_name in current {
+                if !real.remove(&current_name) {
+                    trans.execute("DELETE FROM agents WHERE name = ?1;", &[&current_name])?;
                 }
             }
 
+            // And any *new* agents need to be inserted.
             for missing in &real {
                 trans.execute(
                     "INSERT INTO agents (name) VALUES (?1);",