Nodes the seeder is tracking may not always persist to disk due to races between ThreadDumper and ThreadCrawler #108
Comments
melroy89
pushed a commit
to BitcoinCash1/Bitcoin-Cash-Node
that referenced
this issue
Oct 10, 2024
The app can now be exited in the usual unix ways (SIGTEM, SIGINT, etc). The app now fully cleans itself up on exit, and dumps stats to disk immediately on exit, as opposed to once per hour only (the once per hour thing still occurs in the background as well, however). This issue was also something Sipa wanted to fix, see here: sipa/bitcoin-seeder#76 Note that the extant `ThreadDumper` has an issue when it is saving in the background: sipa/bitcoin-seeder#108. A side-effect of this commit is that when the app exits and it does the final save to disk, no nodes will be lost from the DB, partially resolving the above issue. Summary of changes: - All sleeping in the code (including poll/select) now periodically polls the newly-introduced "shutdown requested" flag. - Added asynch-signal-safe handling of app shutdown signals (SIGINT, etc) - On app signal, set the shutdown requested flag. - ThreadCrawler socket select/poll now periodically wakes up to check the "shutdown requested" flag (this is to allow for fast thread exit from the crawler threads).
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
If
ThreadDumperwakes up to save to disk periodically, it may miss nodes that are otherwise good and won't save them to disk. This is because it basically is getting nodes to save to disk fromCAddrDbfrom its two lists:ourIdandunkId.However, when the crawler threads run, they "take" IPs to check from the
CAddrDbclass. These nodes are temporarily "popped" off the internal listsourIdand/orunkId(only to be returned later once a good/bad result is determined by the crawler).So if
ThreadDumperwakes up while any crawlers are actively checking nodes.. those nodes won't get saved to disk. :'(The text was updated successfully, but these errors were encountered: