Some rocksdb option wrong, storaged upgrade and can't start

[whitewum]

ERROR:

update rocksdb_column_family_options from {"write_buffer_size":"67108864","max_write_buffer_number":"4","max_bytes_for_level_base":"268435456","bloom_locality":"1"} to {"max_background_compactions":"4","level0_file_num_compaction_trigger":"2","bloom_locality":"1","disable_auto_compactions":"false","max_background_jobs":"8","max_bytes_for_level_base":"268435456","max_write_buffer_number":"4","write_buffer_size":"67108864"}

I upgrade storaged version from Feb.20 to Mar 24.

I've update config several times.

I think it's due to rocksdb options check.

[critical27]

That's because some option can't be set dynamically, in your case, "bloom locality" and "write_buffer_size". And for now we just return error. How about these ones take effact when storage restart?

[critical27]

And some option is not the rocksdb_column_family_options, that would make the storage
fail to start when you reboot. You need to find out which is rocksdb_column_family_options and rocksdb_db_options.

[whitewum]

And some option is not the rocksdb_column_family_options, that would make the storage
fail to start when you reboot. You need to find out which is rocksdb_column_family_options and rocksdb_db_options.

Correct. I mix and didn't look carefully at these two kinds of options.

My suggestion is a warning or error when setting config incorrectly. There are too many options in Rocksdb, and error-prone to set them.

It's better than accept the option when setting and fail to restart.

So, you can fix this issue later.

By the way, how to set Rocksdb option are very hard, but very import to improve performance in different on-site cases (more vertex lookup, or more edge scan, or more range deletes). Only one default storage.conf is far from enough for those situations. I prefer the storaged to provide a way to change and check this options. (maybe together with some suggestions)