4 JIT compilation is beneficial primarily for long-running CPU-bound
5 queries. Frequently these will be analytical queries. For short queries
6 the added overhead of performing JIT compilation will often be higher
7 than the time it can save.
9 To determine whether JIT compilation should be used, the total
10 estimated cost of a query (see Chapter 69 and Section 19.7.2) is used.
11 The estimated cost of the query will be compared with the setting of
12 jit_above_cost. If the cost is higher, JIT compilation will be
13 performed. Two further decisions are then needed. Firstly, if the
14 estimated cost is more than the setting of jit_inline_above_cost, short
15 functions and operators used in the query will be inlined. Secondly, if
16 the estimated cost is more than the setting of jit_optimize_above_cost,
17 expensive optimizations are applied to improve the generated code. Each
18 of these options increases the JIT compilation overhead, but can reduce
19 query execution time considerably.
21 These cost-based decisions will be made at plan time, not execution
22 time. This means that when prepared statements are in use, and a
23 generic plan is used (see PREPARE), the values of the configuration
24 parameters in effect at prepare time control the decisions, not the
25 settings at execution time.
29 If jit is set to off, or if no JIT implementation is available (for
30 example because the server was compiled without --with-llvm), JIT will
31 not be performed, even if it would be beneficial based on the above
32 criteria. Setting jit to off has effects at both plan and execution
35 EXPLAIN can be used to see whether JIT is used or not. As an example,
36 here is a query that is not using JIT:
37 =# EXPLAIN ANALYZE SELECT SUM(relpages) FROM pg_class;
39 -------------------------------------------------------------------------------
40 ------------------------------
41 Aggregate (cost=16.27..16.29 rows=1 width=8) (actual time=0.303..0.303 rows=1.
43 Buffers: shared hit=14
44 -> Seq Scan on pg_class (cost=0.00..15.42 rows=342 width=4) (actual time=0.
45 017..0.111 rows=356.00 loops=1)
46 Buffers: shared hit=14
47 Planning Time: 0.116 ms
48 Execution Time: 0.365 ms
50 Given the cost of the plan, it is entirely reasonable that no JIT was
51 used; the cost of JIT would have been bigger than the potential
52 savings. Adjusting the cost limits will lead to JIT use:
53 =# SET jit_above_cost = 10;
55 =# EXPLAIN ANALYZE SELECT SUM(relpages) FROM pg_class;
57 -------------------------------------------------------------------------------
58 ------------------------------
59 Aggregate (cost=16.27..16.29 rows=1 width=8) (actual time=6.049..6.049 rows=1.
61 Buffers: shared hit=14
62 -> Seq Scan on pg_class (cost=0.00..15.42 rows=342 width=4) (actual time=0.
63 019..0.052 rows=356.00 loops=1)
64 Buffers: shared hit=14
65 Planning Time: 0.133 ms
68 Options: Inlining false, Optimization false, Expressions true, Deforming true
69 Timing: Generation 1.259 ms (Deform 0.000 ms), Inlining 0.000 ms, Optimizatio
70 n 0.797 ms, Emission 5.048 ms, Total 7.104 ms
71 Execution Time: 7.416 ms
73 As visible here, JIT was used, but inlining and expensive optimization
74 were not. If jit_inline_above_cost or jit_optimize_above_cost were also
75 lowered, that would change.