Use lightweight query profiling to inspect live operator rows and in-flight plans, while avoiding misleading progress percentages.
NOCOUNT Performance: What SET NOCOUNT ON Saves in a Loop
NOCOUNT performance depends on the messages that SQL Server no longer sends to the client. This post counts the network packets of a procedure with a 5,000 step loop, with the setting off and on. It also explains why some tests show no time gain, and where the setting helps most.
Clone Database in SQL Server With DBCC CLONEDATABASE
Clone Database in SQL Server with DBCC CLONEDATABASE and you copy the structure and the statistics without a single row. This post runs the command on a demo database, checks what the clone holds and what it leaves out, and shows that the optimizer builds the same plan from the copied statistics.
Query Plan Join: Why a Query With No Join Shows One
A query plan join can appear even when the query has no join, because SQL Server joins a nonclustered index to the table itself. This post reads the plan of a one table query, removes the join with a covering index, and finds the row count where SQL Server stops using lookups.
Statistics Sample Percent and PERSIST_SAMPLE_PERCENT
The statistics sample percent decides how many rows SQL Server reads when it updates a statistic. A tested demo on 1,000,000 rows shows the default sample of 35.5 percent, a FULLSCAN of 100 percent, and the way PERSIST_SAMPLE_PERCENT keeps 100 percent for later updates, including an automatic one. It also shows that a sampled update can be slower than a full scan.
APPROX_COUNT_DISTINCT With GROUP BY: When Exact Wins
APPROX_COUNT_DISTINCT with GROUP BY returns an estimate for every group. A tested demo on 2,000,000 rows shows a memory grant of 207,616 KB for the exact count against 3,112 KB for the estimate, and a memory grant of 0 for the exact count once an index exists. It also shows the error on small groups, where 4,648 of 100,000 groups were off.
Row Mode Memory Grant Feedback: It Needs Level 150
Row mode memory grant feedback needs compatibility level 150 or higher, while batch mode needs only 140. A tested demo runs a row mode sort and a batch mode sort at levels 140, 150 and 170. The row sort keeps its grant of 11,888 KB at level 140 and drops to 1,536 KB at 150.







