diff options
| -rw-r--r-- | changes/bug2716 | 5 | ||||
| -rw-r--r-- | src/or/dirserv.c | 22 | ||||
| -rw-r--r-- | src/or/dirserv.h | 12 | ||||
| -rw-r--r-- | src/or/main.c | 7 |
4 files changed, 38 insertions, 8 deletions
diff --git a/changes/bug2716 b/changes/bug2716 new file mode 100644 index 0000000000..4663ed3184 --- /dev/null +++ b/changes/bug2716 @@ -0,0 +1,5 @@ + o Minor features: + - When a relay has failed several reachability tests, be more accurate + at recording when it became unreachable, so we can in turn provide + more accuracy at assigning Stable, Guard, HSDir, etc flags. Bugfix + on 0.2.0.6-alpha. Resolves bug 2716. diff --git a/src/or/dirserv.c b/src/or/dirserv.c index cbf8c3685a..1796c28418 100644 --- a/src/or/dirserv.c +++ b/src/or/dirserv.c @@ -994,8 +994,18 @@ dirserv_set_router_is_running(routerinfo_t *router, time_t now) } if (!answer && running_long_enough_to_decide_unreachable()) { - /* not considered reachable. tell rephist. */ - rep_hist_note_router_unreachable(router->cache_info.identity_digest, now); + /* Not considered reachable. tell rephist about that. + + Because we launch a reachability test for each router every + REACHABILITY_TEST_CYCLE_PERIOD seconds, then the router has probably + been down since at least that time after we last successfully reached + it. + */ + time_t when = now; + if (router->last_reachable && + router->last_reachable + REACHABILITY_TEST_CYCLE_PERIOD < now) + when = router->last_reachable + REACHABILITY_TEST_CYCLE_PERIOD; + rep_hist_note_router_unreachable(router->cache_info.identity_digest, when); } node->is_running = answer; @@ -3232,8 +3242,8 @@ dirserv_single_reachability_test(time_t now, routerinfo_t *router) * try a few connections per call. * * The load balancing is such that if we get called once every ten - * seconds, we will cycle through all the tests in 1280 seconds (a - * bit over 20 minutes). + * seconds, we will cycle through all the tests in + * REACHABILITY_TEST_CYCLE_PERIOD seconds (a bit over 20 minutes). */ void dirserv_test_reachability(time_t now) @@ -3259,11 +3269,11 @@ dirserv_test_reachability(time_t now) continue; /* bridge authorities only test reachability on bridges */ // if (router->cache_info.published_on > cutoff) // continue; - if ((((uint8_t)id_digest[0]) % 128) == ctr) { + if ((((uint8_t)id_digest[0]) % REACHABILITY_MODULO_PER_TEST) == ctr) { dirserv_single_reachability_test(now, router); } } SMARTLIST_FOREACH_END(router); - ctr = (ctr + 1) % 128; /* increment ctr */ + ctr = (ctr + 1) % REACHABILITY_MODULO_PER_TEST; /* increment ctr */ } /** Given a fingerprint <b>fp</b> which is either set if we're looking for a diff --git a/src/or/dirserv.h b/src/or/dirserv.h index 0489f3a6a8..8bd6d18f96 100644 --- a/src/or/dirserv.h +++ b/src/or/dirserv.h @@ -12,6 +12,18 @@ #ifndef _TOR_DIRSERV_H #define _TOR_DIRSERV_H +/** What fraction (1 over this number) of the relay ID space do we + * (as a directory authority) launch connections to at each reachability + * test? */ +#define REACHABILITY_MODULO_PER_TEST 128 + +/** How often (in seconds) do we launch reachability tests? */ +#define REACHABILITY_TEST_INTERVAL 10 + +/** How many seconds apart are the reachability tests for a given relay? */ +#define REACHABILITY_TEST_CYCLE_PERIOD \ + (REACHABILITY_TEST_INTERVAL*REACHABILITY_MODULO_PER_TEST) + /** Maximum length of an exit policy summary. */ #define MAX_EXITPOLICY_SUMMARY_LEN 1000 diff --git a/src/or/main.c b/src/or/main.c index c1335ff5d9..9c19485990 100644 --- a/src/or/main.c +++ b/src/or/main.c @@ -1049,6 +1049,7 @@ run_scheduled_events(time_t now) static time_t time_to_write_stats_files = 0; static time_t time_to_write_bridge_stats = 0; static time_t time_to_check_port_forwarding = 0; + static time_t time_to_launch_reachability_tests = 0; static int should_init_bridge_stats = 1; static time_t time_to_retry_dns_init = 0; static time_t time_to_next_heartbeat = 0; @@ -1151,8 +1152,10 @@ run_scheduled_events(time_t now) if (accounting_is_enabled(options)) accounting_run_housekeeping(now); - if (now % 10 == 0 && (authdir_mode_tests_reachability(options)) && - !we_are_hibernating()) { + if (time_to_launch_reachability_tests < now && + (authdir_mode_tests_reachability(options)) && + !we_are_hibernating()) { + time_to_launch_reachability_tests = now + REACHABILITY_TEST_INTERVAL; /* try to determine reachability of the other Tor relays */ dirserv_test_reachability(now); } |