relays checkpoint their state file twice a day

2 files changed, 15 insertions, 1 deletions

diff --git a/changes/bug3012 b/changes/bug3012
new file mode 100644
index 0000000000..dfde5fa90c
--- /dev/null
+++ b/changes/bug3012
@@ -0,0 +1,5 @@
+  o Minor features:
+    - Relays can go for weeks without writing out their state file. A
+      relay that crashes would lose its bandwidth history (including
+      capacity estimate), client country statistics, and so on. Now relays
+      checkpoint the file at least every 12 hours. Addresses bug 3012.
diff --git a/src/or/config.c b/src/or/config.c
index 9384b3a68a..dc24140483 100644
--- a/src/or/config.c
+++ b/src/or/config.c
@@ -5122,6 +5122,11 @@ or_state_load(void)
 /** If writing the state to disk fails, try again after this many seconds. */
 #define STATE_WRITE_RETRY_INTERVAL 3600
 
+/** If we're a relay, how often should we checkpoint our state file even
+ * if nothing else dirties it? This will checkpoint ongoing stats like
+ * bandwidth used, per-country user stats, etc. */
+#define STATE_RELAY_CHECKPOINT_INTERVAL (12*60*60)
+
 /** Write the persistent state to disk. Return 0 for success, <0 on failure. */
 int
 or_state_save(time_t now)
@@ -5172,7 +5177,11 @@ or_state_save(time_t now)
   tor_free(fname);
   tor_free(contents);
 
-  global_state->next_write = TIME_MAX;
+  if (server_mode(get_options()))
+    global_state->next_write = now + STATE_RELAY_CHECKPOINT_INTERVAL;
+  else
+    global_state->next_write = TIME_MAX;
+
   return 0;
 }