SQLite table disk usage

How can I find out the disk usage of a single table inside a SQLite database without copying it in a new empty database?

You can use sqlite3_analyzer from https://www.sqlite.org/download.html.

It's a really cool tool. It shows you the number of pages used by each table with and without indexes (each page, by default, is 1024 bytes).

This is a sample sqlite3_analyzer output for the Northwind database:

*** Page counts for all tables with their indices ********************

EMPLOYEES............................. 200         34.4% 
ORDERS................................ 152         26.2% 
CATEGORIES............................ 90          15.5% 
ORDER DETAILS......................... 81          13.9% 
CUSTOMERS............................. 17           2.9% 
SQLITE_MASTER......................... 11           1.9% 
PRODUCTS.............................. 7            1.2% 
SUPPLIERS............................. 7            1.2% 
TERRITORIES........................... 6            1.0% 
CUSTOMERCUSTOMERDEMO.................. 2            0.34% 
CUSTOMERDEMOGRAPHICS.................. 2            0.34% 
EMPLOYEETERRITORIES................... 2            0.34% 
REGION................................ 2            0.34% 
SHIPPERS.............................. 2            0.34% 

It also generates SQL statements which can be used to create a database with usage statistics, which you can then analyze.

If you are on linux or OSX, or otherwise have the unix utilities awk (and optionally, sort) available, you can do the following to get counts and estimated size via dump analysis:

# substitute '.dump' for '.dump mytable' if you want to limit to specific table
sqlite3 db.sqlite3 '.dump' | awk -f sqlite3_size.awk

which returns:

table            count   est. size
my_biggest_table 1090    60733958
my_table2        26919   7796902
my_table3        10390   2732068

and uses awk script:

/INSERT INTO/ {                              # parse INSERT commands
    split($0, values, "VALUES");             # extract everything after VALUES
    split(values[1], name, "INSERT INTO");   # get tablename
    tablename = name[2];                     #
    gsub(/[\047\042]/, "", tablename);         # remove single and double quotes from name
    gsub(/[\047,]/, "", values[2]);          # remove single-quotes and commas
    sizes[tablename] += length(values[2]) - 3; # subtract 3 for parens and semicolon
    counts[tablename] += 1;
}

END {
    print "table\tcount\test. size"
    for(k in sizes) {
        # print and sort in descending order:
        print k "\t" counts[k] "\t" sizes[k] | "sort -k3 -n -r";

        # or, if you don't have the sort command:
        print k "\t" counts[k] "\t" sizes[k];
    }
}

The estimated size is based on the string length of the "INSERT INTO" command, and so is not going to equal the actual size on disk, but for me, count plus the estimated size is more useful than other alternatives such as page count.