Explicitly handle BDB::Fatal errors by dumping database stats and restarting

2 files changed, 27 insertions, 3 deletions

diff --git a/lib/rbot/dbhash.rb b/lib/rbot/dbhash.rb
index 3168f208..2170ea08 100644
--- a/lib/rbot/dbhash.rb
+++ b/lib/rbot/dbhash.rb
@@ -135,6 +135,21 @@ module Irc
       end
     end
 
+    def DBTree.stats()
+      begin
+        debug "General stats:"
+        debug @@env.stat
+        debug "Lock stats:"
+        debug @@env.lock_stat
+        debug "Log stats:"
+        debug @@env.log_stat
+        debug "Txn stats:"
+        debug @@env.txn_stat
+      rescue
+        debug "Couldn't dump stats"
+      end
+    end
+
     def DBTree.cleanup_env()
       begin
         debug "DBTree: checking transactions ..."
diff --git a/lib/rbot/ircbot.rb b/lib/rbot/ircbot.rb
index d513cb71..8e6fabac 100644
--- a/lib/rbot/ircbot.rb
+++ b/lib/rbot/ircbot.rb
@@ -376,11 +376,19 @@ class IrcBot
       rescue SystemExit
         exit 0
       rescue TimeoutError, SocketError => e
-        puts "#{e.class}: #{e}"
+        puts "network exception: #{e.class}: #{e}"
+        debug e.inspect
         debug e.backtrace.join("\n")
+      rescue BDB::Fatal => e
+        puts "fatal bdb error: #{e.class}: #{e}"
+        debug e.inspect
+        debug e.backtrace.join("\n")
+        DBTree.stats
+        restart("Oops, we seem to have registry problems ...")
       rescue Exception => e
-        puts "network exception: #{e.inspect}"
-        puts e.backtrace.join("\n")
+        puts "non-net exception: #{e.class}: #{e}"
+        debug e.inspect
+        debug e.backtrace.join("\n")
         @socket.shutdown # now we reconnect
       rescue => e
         puts "unexpected exception: connection closed: #{e.inspect}"
@@ -545,6 +553,7 @@ class IrcBot
     shutdown(msg)
     sleep @config['server.reconnect_wait']
     # now we re-exec
+    # Note, this fails on Windows
     exec($0, *@argv)
   end