[CASSANDRA-13027] Questions on compaction process

• Type: Task
• Status: Resolved
• Priority: Normal
• Resolution: Invalid
• Fix Version/s: 2.2.5
• Component/s: Local/Compaction

Environment

ubuntu, centos

• Assignee: Unassigned
• Reporter: Arvind Viswanathan
• Created: 09/Dec/16 19:25
• Updated: 16/Apr/19 09:30
• Resolved: 12/Dec/16 08:32

Description

I have the following schema ( a raw table and index table to look up data from the raw table) and I notice the following problem:

create table raw_table(
key uuid PRIMARY KEY,
value text) ;

create table index_table_1(
time int,
index_field1 text,
col1 text,
value uuid); PRIMARY_KEY(int, text)

My usage involves continuous writes( no updates, just new data with TTL of 48 hrs) and very less reads. I use STCS. I noticed the reads to the raw_table are failing whereas the one to index tables where succeeding. I took a look at the compaction history of the different table and here is what I observe

Raw table has low completedcompactions (from the nodetool compactionhistory) but high number of SSTable count (from nodetool cfstats), whereas the index table has lot of completed compactions and low SSTable count. I can understand that high SSTable count can lead to read failures, but I don't understand why the compactions are not able to complete as fast in the raw tables. The cassandra version I use is 2.2.5.
Also what is the impact of compaction on append only tables?