mikedilger on Nostr: GOSSIP USERS: Gossip is now using LMDB! (master branch). Here is what this means: - ...
GOSSIP USERS: Gossip is now using LMDB! (master branch). Here is what this means:
- Your client should be faster
- Your client should consume less disk space. In fact, because we aren't deleting the gossip.sqlite file, it will consume strictly more disk space. I recommend you keep the gossip.sqlite file for a while just in case, but delete it eventually. Also note that LMDB requires us to specify a maximum database size, and it reserves a sparse file of that size which appears to be 128 GB. But it isn't really. A 'du' command will show how much space it really uses.
- If your client crashes, you lose power, or you ^C to exit, there is a chance that the last things written will not have been flushed to disk. This shouldn't affect integrity, we should have all of ACID except the D. However, if your filesystem does not preserve write order, all bets are off.
- A number of bugs around missing events have been resolved.
- The UI may be LESS responsive. I will continue to work to improve that situation.
- I recommend using an SSD if you can (however I'm using a spinning disk and it works fine for me)
The LMDB work is not yet done. I will be migrating from lmdb-rkv to heed. I will also be doing migrations that change the indexes to get better performance. But other non-lmdb work can now again progress as usual.
Let me know if there are problems, preferably on github.
- Your client should be faster
- Your client should consume less disk space. In fact, because we aren't deleting the gossip.sqlite file, it will consume strictly more disk space. I recommend you keep the gossip.sqlite file for a while just in case, but delete it eventually. Also note that LMDB requires us to specify a maximum database size, and it reserves a sparse file of that size which appears to be 128 GB. But it isn't really. A 'du' command will show how much space it really uses.
- If your client crashes, you lose power, or you ^C to exit, there is a chance that the last things written will not have been flushed to disk. This shouldn't affect integrity, we should have all of ACID except the D. However, if your filesystem does not preserve write order, all bets are off.
- A number of bugs around missing events have been resolved.
- The UI may be LESS responsive. I will continue to work to improve that situation.
- I recommend using an SSD if you can (however I'm using a spinning disk and it works fine for me)
The LMDB work is not yet done. I will be migrating from lmdb-rkv to heed. I will also be doing migrations that change the indexes to get better performance. But other non-lmdb work can now again progress as usual.
Let me know if there are problems, preferably on github.