Copy and Delete Data Across Tables When There is no Space
Recently I needed to apply compression data on a particularly large table. One of the main reasons for applying compression was because the database was extremely low on space, in both the data and the log files. To make matters worse, the data and log files were nowhere near big enough to accommodate compressing the entire table in one go. If the able was partitioned then I could have done one partition at a time and all my problems would go away. No such luck.
So I had to think quite outside of the box to compressing the data. The solution I came up with was to create a new empty table, and copy 50 rows at a time (that was the largest I could move at any one time, yes I really was THAT low on space) and then delete the 50 rows copied. But instead of COPY/DELETE, which would seem the logical way, I ran the DELETE first and then run an INSERT from the output table. This way I would be certain that the same rows that were deleted were copied into the new table, and meant that I was only scanning the original table the once.
I had to run further checks into the script: the database was set to simple mode, but the log file would still grow because of ETL processes etc running on other parts of the db, and so I needed to check to make sure that I wasn’t saturating the log file with each delete. So the logic here will work for both SIMPLE, BULK and FULL recovery models.