yiwu-arbug
yiwu-arbug
By default each cache shard is at least 512KB, which means if cache size is set to 0, there's only 1 shard. It will of course cause mutex contention. Can...
yes, we saw similar issue when enabling dictionary compression on vanilla RocksDB. we end up limit the sample size to 8MB per SST. cc @hunterlxt
@DorianZheng Long time no see! Yeah, we should probably do what you suggest.
> @yiwu-arbug Thanks for the kindness. > @Connor1996 > > ``` > How about just recording the sequence number after rewriting index, and mark the blob file as deleted until...
@DorianZheng Thanks. Though I think we can access rocksdb internals from Titan directly instead of adding new rocksdb API.
I don't know how to fix this one, since `OnFlushCompleted` don't know what's the input data size for each blob file.
@wujy-cs sounds reasonable. So basically, a) mark the latest memtable id (say, M) after GC finish, b) wait till M flushed, then subtract file size by sum of output size...
Indeed we would need to double check `HasBGError()` while holding `mutex_` before every time calling `LogAndApply()`. Do you mind sending a fix?
> Actually I prefer the former implementation of `Version` of blob files which can exploit the advantage of `MVCC` notion and it will downgrade to current implementation only if read...
> I built the titan on the version in the demo, with rocksdb version 6.4.tikv. > > > ./titandb_bench --db="/dev/shm/titan_bench" --use_existing_db=false --titan_min_blob_size=0 --value_size=1024 --num=10000000 --compression_type=none --benchmarks="fillseq" > > Then I...