Home/ Questions/Q 6853467
In Process

The Archive Base Latest Questions

Editorial Team
  • 0
Asked: 2026-05-27T01:30:55+00:00

I have a sqlite db on an ARM embedded platform running Linux with somewhat

  • 0

I have a sqlite db on an ARM embedded platform running Linux with somewhat limited resources. Storage device is a microSD card. Sqlite version is 3.7.7.1. The application accessing sqlite is written in C++.

I want to know the number of rows in several tables in regular intervals. I currently use

select count(*) from TABLENAME;

to get this information. I’m having trouble with the performance: When the table sizes reach a certain point (~200K lines), I have a lot of system and iowait load every time I check the table sizes.

When I wrote this, I though looking up the number of rows in a table would be fast as it is probably stored somewhere. But now I’m suspecting that sqlite actually looks through all rows and when I pass the point where the data doesn’t fit into the disk cache anymore I get a lot of io load. This would roughly fit from db size and available memory.

Can anyone tell me if sqlite behaves in the way I suspect?

Is there any way to get the number of table rows without producing this amount of load?

EDIT: plaes has asked about the table layout:

CREATE TABLE %s (timestamp INTEGER PRIMARY KEY, offset INTEGER, value NUMERIC);
Share

Leave an answer

You must login to add an answer.

Need An Account,

1 Answer

  1. Editorial Team

    From all the information I gathered, count() apparently really needs to scan the table. As plaes has pointed out, this is faster if the count is done on a integer indexed column, but scanning the index is still needed.

    What I do now is store the row count somewhere and increment / decrement it manually in the same transactions I use to do inserts and deletes to keep it consistent.

    • 0