mwish

Results 249 comments of mwish
trafficstars

@rok I've written a bit ugly random string generator: ```C++ std::shared_ptr generateShareSeqString( ::arrow::random::RandomArrayGenerator& gen, size_t seqNum, double nullPercent) { ::arrow::StringBuilder builder; auto seed = gen.seed(); std::shared_ptr nullBitMap = nullptr; if...

Perhaps, I don't know. IMO, Prefix filter is good for range scan, however, Range in kvrocks is not so widely used...

https://github.com/tikv/tikv/blob/2472fd4d85c220b74dc889b493e70bc95dcc75c2/src/config/mod.rs#L963 TiKV uses prefix bf to remove the timestamp column. I think if we'd like to use prefix bf, maybe we need a well-defined rule for prefix. Little commands uses...

You can first draft a format in discussion, and then move forward to implement it. Some database uses HLL to optimize the ndv: 1. https://github.com/facebookincubator/velox/tree/main/velox/common/hyperloglog 2. https://github.com/ClickHouse/ClickHouse/blob/92e63bf45c3cd36163131e42d4bd003e543031f4/src/Common/HyperLogLogCounter.h You can choose...

Initialize S3/GCS etc would be a bit tricky, maybe opendal C SDK would help: https://github.com/apache/incubator-opendal . It would be also ok for testing in local machine. Other tools in C++...

🤔 ClickHouse can read from remote S3, so I think it's able to upload or backup to s3. However, TiKV only supports a br here. (See: https://tikv.org/docs/6.5/concepts/explore-tikv-features/backup-restore-cn/ ). Maybe we...

@Xuanwo Here I think the performance is not the critical reason and we may not enable some advance feature about threading, I think opendal as a backend of RocksDB Env...

After some discussion, maybe design some new syntax and using another thread / process to upload Backup in Local FileSystem to HDFS/S3 is also a way. This avoid the complex...