search

jlongster / absurd-sql

sqlite3 in ur indexeddb (hopefully a better backend soon)
MIT License
4.15k stars 101 forks source link

Fix bulkSet #56

Open quolpr opened 2 years ago

quolpr commented 2 years ago

~I think it may potentially fix https://github.com/jlongster/absurd-sql/issues/30~

After that fix, the writing performance become slower. But we should await before all the writes finish to finish the transaction. Otherwise, the next read IDB (when you want to execute SELECT for example) will need to await when previous INSERT will finish, which is not fair for the SELECT query. And, it seems the benchmarks are wrong at the article (but not so much! The write performance is still amazing).

Also, I don't see any performance penalty of Promise.all usage — the new IDB transactions are not spawning, the overall performance at the first glance is the same.

quolpr commented 2 years ago

Here is the example, without await and with. You can notice that without await the next select query slow down a lot (cause IDB wait write transaction to finish). And with await, the next select query is fast. And in sum, the run time for both cases is 2 secs.

https://user-images.githubusercontent.com/7958527/179366981-4d1819b3-1074-4d84-b5de-6b92513ab0c1.mp4

quolpr commented 2 years ago

Update: this PR actually doesn't fix #30 😢 But still without awaiting the next query will be much slower, cause internally IDB don't finish writing and next read will await those writings to finish