Loading…
Mark's stack trace: https://gist.github.com/mdcallag/152f8631210a7cd3c0dd
Currently drop_index_thread calls Seek() to check if the target index is was completely removed or not, and during that time a global mutex drop_index_interrupte_mutex is held, which blocks delete_table() which is called via DROP TABLE -- https://github.com/MySQLOnRocksDB/mysql-5.6/blob/webscalesql-5.6.23.71/storage/rocksdb/ha_rocksdb.cc#L5098
We know that Seek() may take seconds if there were huge number of tombstones. This means current implementation of DROP TABLE may also take seconds. This has to be fixed.
https://reviews.facebook.net/D44985
Mark's stack trace: https://gist.github.com/mdcallag/152f8631210a7cd3c0dd
Currently drop_index_thread calls Seek() to check if the target index is was completely removed or not, and during that time a global mutex drop_index_interrupte_mutex is held, which blocks delete_table() which is called via DROP TABLE -- https://github.com/MySQLOnRocksDB/mysql-5.6/blob/webscalesql-5.6.23.71/storage/rocksdb/ha_rocksdb.cc#L5098
We know that Seek() may take seconds if there were huge number of tombstones. This means current implementation of DROP TABLE may also take seconds. This has to be fixed.