我正在比较 9.1.9/9.3.6 和 9.4.1。我认为9.4.1的性能应该更好或者至少不差。 但我这里有一个问题。 我已经在同一台机器上安装了两个数据库，恢复了相同的转储，使用了 ANALYZE，我得到了 2 个结果，其中我发现由于某种原因 9.1.9 运行得更快 - 118ms vs 452ms。 我使用 EXPLAIN ANALYZE 检查了相同的 sql 查询(具有相同的 work_mem = 8MB 和 share_buffers = 128MB)。
1) 9.1.9
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------------------------
Limit (cost=6929.93..6929.94 rows=1 width=10) (actual time=99.175..99.178 rows=25 loops=1)
-> Sort (cost=6929.93..6929.94 rows=1 width=10) (actual time=99.175..99.175 rows=25 loops=1)
Sort Key: (CASE WHEN (p.f_without_price = 1) THEN 0::double precision ELSE p1.price END)
Sort Method: top-N heapsort Memory: 26kB
-> Nested Loop Anti Join (cost=133.65..6929.92 rows=1 width=10) (actual time=0.345..97.884 rows=4876 loops=1)
Join Filter: ((p2.price < p1.price) OR ((p2.price = p1.price) AND (p2.id < p1.id)))
-> Nested Loop Left Join (cost=90.19..6882.40 rows=1 width=64) (actual time=0.344..64.038 rows=4891 loops=1)
Filter: (((p.f_without_price = 1) OR (p1.price <> 0::double precision)) AND (CASE WHEN (p.f_without_price = 1) THEN 0::double precision ELSE p1.price END <> 0::double precision))
-> Nested Loop Left Join (cost=46.75..6834.91 rows=1 width=48) (actual time=0.344..23.213 rows=4889 loops=1)
Join Filter: (p.f_assignable = 0)
Filter: (((p.f_assignable = 1) AND (pc.product_id IS NOT NULL)) OR ((p.f_assignable = 0) AND (pch.product_id IS NULL)))
-> Nested Loop Left Join (cost=46.75..6822.71 rows=2 width=58) (actual time=0.344..16.257 rows=4907 loops=1)
Join Filter: (p.f_assignable = 1)
-> Nested Loop (cost=46.75..6809.23 rows=2 width=54) (actual time=0.343..8.678 rows=4907 loops=1)
-> Nested Loop (cost=0.00..24.84 rows=1 width=58) (actual time=0.017..0.052 rows=2 loops=1)
-> Nested Loop (cost=0.00..16.55 rows=1 width=54) (actual time=0.013..0.035 rows=3 loops=1)
Join Filter: (cs.supplier_id = grps.supplier_id)
-> Index Scan using cs_aggregated_groups_pkey on cs_aggregated_groups grps (cost=0.00..8.27 rows=1 width=50) (actual time=0.006..0.008 rows=3 loops=1)
Index Cond: (customer_id = 1388)
-> Index Scan using index_15 on cs_groups cs (cost=0.00..8.27 rows=1 width=8) (actual time=0.004..0.007 rows=3 loops=3)
Index Cond: (customer_id = 1388)
Filter: (f_primary_group = 1)
-> Index Scan using index_26 on supplier s (cost=0.00..8.27 rows=1 width=4) (actual time=0.004..0.004 rows=1 loops=3)
Index Cond: (id = cs.supplier_id)
Filter: ((f_active = 1) AND (f_deleted = 0))
-> Bitmap Heap Scan on product p (cost=46.75..6754.19 rows=2416 width=12) (actual time=0.401..3.683 rows=2454 loops=2)
Recheck Cond: (supplier_id = s.id)
Filter: ((f_available = 1) AND (f_active = 1) AND (f_deleted = 0) AND ((f_assignable = 1) OR (f_assignable = 0)))
-> Bitmap Index Scan on index_57 (cost=0.00..46.41 rows=2416 width=0) (actual time=0.320..0.320 rows=3550 loops=2)
Index Cond: (supplier_id = s.id)
-> Index Scan using index_49 on product_customer pc (cost=0.00..6.72 rows=1 width=8) (actual time=0.001..0.001 rows=0 loops=4907)
Index Cond: ((product_id = p.id) AND (customer_id = 1388) AND (customer_id = cs.customer_id))
-> Index Scan using index_63 on product_customer_hidden pch (cost=0.00..6.08 rows=1 width=8) (actual time=0.001..0.001 rows=0 loops=4907)
Index Cond: ((product_id = p.id) AND (customer_id = 1388) AND (customer_id = cs.customer_id))
-> Bitmap Heap Scan on price p1 (cost=43.44..47.46 rows=1 width=27) (actual time=0.007..0.007 rows=1 loops=4889)
Recheck Cond: ((product_id = p.id) AND ((group_id)::text = ANY ((grps.grp_ids)::text[])) AND (amount = 1))
Filter: (f_valid = 1)
-> Bitmap Index Scan on price_product_id_group_id_amount_type_id_valid_from_valid_t_key (cost=0.00..43.44 rows=1 width=0) (actual time=0.006..0.006 rows=1 loops=4889)
Index Cond: ((product_id = p.id) AND ((group_id)::text = ANY ((grps.grp_ids)::text[])) AND (amount = 1))
-> Bitmap Heap Scan on price p2 (cost=43.47..47.48 rows=1 width=27) (actual time=0.006..0.006 rows=1 loops=4891)
Recheck Cond: ((product_id = p1.product_id) AND ((group_id)::text = ANY ((grps.grp_ids)::text[])) AND (amount = 1) AND (amount = p1.amount))
Filter: (f_valid = 1)
-> Bitmap Index Scan on price_product_id_group_id_amount_type_id_valid_from_valid_t_key (cost=0.00..43.47 rows=1 width=0) (actual time=0.005..0.005 rows=1 loops=4891)
Index Cond: ((product_id = p1.product_id) AND ((group_id)::text = ANY ((grps.grp_ids)::text[])) AND (amount = 1) AND (amount = p1.amount))
Total runtime: 99.297 ms
(45 rows)
2) 9.4.1
QUERY PLAN
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Limit (cost=503.71..503.72 rows=1 width=10) (actual time=416.762..416.764 rows=25 loops=1)
-> Sort (cost=503.71..503.72 rows=1 width=10) (actual time=416.734..416.734 rows=25 loops=1)
Sort Key: (CASE WHEN (p.f_without_price = 1) THEN 0::double precision ELSE p1.price END)
Sort Method: top-N heapsort Memory: 26kB
-> Nested Loop Anti Join (cost=37.73..503.70 rows=1 width=10) (actual time=0.736..415.337 rows=4876 loops=1)
-> Nested Loop Left Join (cost=37.16..497.90 rows=1 width=64) (actual time=0.697..379.099 rows=4891 loops=1)
Join Filter: ((p1.group_id)::text = ANY ((grps.grp_ids)::text[]))
Rows Removed by Join Filter: 485609
Filter: (((p.f_without_price = 1) OR (p1.price <> 0::double precision)) AND (CASE WHEN (p.f_without_price = 1) THEN 0::double precision ELSE p1.price END <> 0::double precision))
Rows Removed by Filter: 13
-> Nested Loop Left Join (cost=36.61..487.14 rows=1 width=48) (actual time=0.619..24.466 rows=4889 loops=1)
Join Filter: ((p.f_assignable = 0) AND (pch.customer_id = cs.customer_id))
Filter: (((p.f_assignable = 1) AND (pc.product_id IS NOT NULL)) OR ((p.f_assignable = 0) AND (pch.product_id IS NULL)))
Rows Removed by Filter: 18
-> Nested Loop Left Join (cost=36.33..486.50 rows=2 width=58) (actual time=0.595..17.019 rows=4907 loops=1)
Join Filter: ((p.f_assignable = 1) AND (pc.customer_id = cs.customer_id))
-> Nested Loop (cost=36.03..485.82 rows=2 width=54) (actual time=0.573..8.876 rows=4907 loops=1)
-> Nested Loop (cost=0.71..18.42 rows=1 width=58) (actual time=0.101..0.122 rows=2 loops=1)
-> Nested Loop (cost=0.43..16.49 rows=1 width=12) (actual time=0.078..0.091 rows=2 loops=1)
-> Index Scan using index_15 on cs_groups cs (cost=0.28..8.30 rows=1 width=8) (actual time=0.020..0.024 rows=3 loops=1)
Index Cond: (customer_id = 1388)
Filter: (f_primary_group = 1)
-> Index Scan using index_26 on supplier s (cost=0.15..8.17 rows=1 width=4) (actual time=0.006..0.007 rows=1 loops=3)
Index Cond: (id = cs.supplier_id)
Filter: ((f_active = 1) AND (f_deleted = 0))
Rows Removed by Filter: 0
-> Index Scan using cs_aggregated_groups_pkey on cs_aggregated_groups grps (cost=0.28..1.92 rows=1 width=50) (actual time=0.004..0.005 rows=1 loops=2)
Index Cond: ((customer_id = 1388) AND (supplier_id = s.id))
-> Bitmap Heap Scan on product p (cost=35.32..454.81 rows=1259 width=12) (actual time=0.456..3.332 rows=2454 loops=2)
Recheck Cond: (supplier_id = s.id)
Filter: ((f_available = 1) AND (f_active = 1) AND (f_deleted = 0))
Rows Removed by Filter: 1096
Heap Blocks: exact=2866
-> Bitmap Index Scan on index_57 (cost=0.00..35.01 rows=2274 width=0) (actual time=0.300..0.300 rows=3550 loops=2)
Index Cond: (supplier_id = s.id)
-> Index Only Scan using index_49 on product_customer pc (cost=0.29..0.33 rows=1 width=8) (actual time=0.001..0.001 rows=0 loops=4907)
Index Cond: ((product_id = p.id) AND (customer_id = 1388))
Heap Fetches: 0
-> Index Only Scan using index_63 on product_customer_hidden pch (cost=0.28..0.30 rows=1 width=8) (actual time=0.001..0.001 rows=0 loops=4907)
Index Cond: ((product_id = p.id) AND (customer_id = 1388))
Heap Fetches: 0
-> Index Scan using price_product_id_group_id_amount_type_id_valid_from_valid_t_key on price p1 (cost=0.56..5.36 rows=166 width=27) (actual time=0.006..0.056 rows=100 loops=4889)
Index Cond: ((product_id = p.id) AND (amount = 1))
Filter: (f_valid = 1)
Rows Removed by Filter: 0
-> Index Scan using price_product_id_group_id_amount_type_id_valid_from_valid_t_key on price p2 (cost=0.56..5.79 rows=1 width=27) (actual time=0.006..0.006 rows=0 loops=4891)
Index Cond: ((product_id = p1.product_id) AND ((group_id)::text = ANY ((grps.grp_ids)::text[])) AND (amount = p1.amount) AND (amount = 1))
Filter: ((f_valid = 1) AND ((price < p1.price) OR ((price = p1.price) AND (id < p1.id))))
Rows Removed by Filter: 1
Planning time: 2.975 ms
Execution time: 416.935 ms
(51 rows)
3) 9.3.6
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Limit (cost=5705.15..5705.16 rows=1 width=10) (actual time=101.671..101.674 rows=25 loops=1)
-> Sort (cost=5705.15..5705.16 rows=1 width=10) (actual time=101.667..101.669 rows=25 loops=1)
Sort Key: (CASE WHEN (p.f_without_price = 1) THEN 0::double precision ELSE p1.price END)
Sort Method: top-N heapsort Memory: 26kB
-> Nested Loop Anti Join (cost=42.85..5705.14 rows=1 width=10) (actual time=0.505..100.493 rows=4876 loops=1)
-> Nested Loop Left Join (cost=42.29..5662.50 rows=1 width=64) (actual time=0.502..68.448 rows=4891 loops=1)
Filter: (((p.f_without_price = 1) OR (p1.price <> 0::double precision)) AND (CASE WHEN (p.f_without_price = 1) THEN 0::double precision ELSE p1.price END <> 0::double precision))
Rows Removed by Filter: 13
-> Nested Loop Left Join (cost=41.72..5612.79 rows=1 width=48) (actual time=0.499..26.178 rows=4889 loops=1)
Join Filter: ((p.f_assignable = 0) AND (pch.customer_id = cs.customer_id))
Filter: (((p.f_assignable = 1) AND (pc.product_id IS NOT NULL)) OR ((p.f_assignable = 0) AND (pch.product_id IS NULL)))
Rows Removed by Filter: 18
-> Nested Loop Left Join (cost=41.44..5600.55 rows=2 width=58) (actual time=0.497..19.018 rows=4907 loops=1)
Join Filter: ((p.f_assignable = 1) AND (pc.customer_id = cs.customer_id))
-> Nested Loop (cost=41.15..5587.11 rows=2 width=54) (actual time=0.495..11.064 rows=4907 loops=1)
-> Nested Loop (cost=0.71..23.07 rows=1 width=58) (actual time=0.024..0.049 rows=2 loops=1)
-> Nested Loop (cost=0.43..16.49 rows=1 width=12) (actual time=0.017..0.035 rows=2 loops=1)
-> Index Scan using index_15 on cs_groups cs (cost=0.28..8.30 rows=1 width=8) (actual time=0.011..0.016 rows=3 loops=1)
Index Cond: (customer_id = 1388)
Filter: (f_primary_group = 1)
-> Index Scan using index_26 on supplier s (cost=0.15..8.17 rows=1 width=4) (actual time=0.003..0.004 rows=1 loops=3)
Index Cond: (id = cs.supplier_id)
Filter: ((f_active = 1) AND (f_deleted = 0))
Rows Removed by Filter: 0
-> Index Scan using cs_aggregated_groups_pkey on cs_aggregated_groups grps (cost=0.28..6.58 rows=1 width=50) (actual time=0.004..0.005 rows=1 loops=2)
Index Cond: ((customer_id = 1388) AND (supplier_id = s.id))
-> Bitmap Heap Scan on product p (cost=40.44..5551.89 rows=1215 width=12) (actual time=0.471..4.755 rows=2454 loops=2)
Recheck Cond: (supplier_id = s.id)
Filter: ((f_available = 1) AND (f_active = 1) AND (f_deleted = 0) AND ((f_assignable = 1) OR (f_assignable = 0)))
Rows Removed by Filter: 1096
-> Bitmap Index Scan on index_57 (cost=0.00..40.14 rows=2235 width=0) (actual time=0.326..0.326 rows=3550 loops=2)
Index Cond: (supplier_id = s.id)
-> Index Only Scan using index_49 on product_customer pc (cost=0.29..6.70 rows=1 width=8) (actual time=0.001..0.001 rows=0 loops=4907)
Index Cond: ((product_id = p.id) AND (customer_id = 1388))
Heap Fetches: 0
-> Index Only Scan using index_63 on product_customer_hidden pch (cost=0.28..6.10 rows=1 width=8) (actual time=0.001..0.001 rows=0 loops=4907)
Index Cond: ((product_id = p.id) AND (customer_id = 1388))
Heap Fetches: 0
-> Index Scan using price_product_id_group_id_amount_type_id_valid_from_valid_t_key on price p1 (cost=0.56..49.70 rows=1 width=27) (actual time=0.007..0.007 rows=1 loops=4889)
Index Cond: ((product_id = p.id) AND ((group_id)::text = ANY ((grps.grp_ids)::text[])) AND (amount = 1))
Filter: (f_valid = 1)
Rows Removed by Filter: 0
-> Index Scan using price_product_id_group_id_amount_type_id_valid_from_valid_t_key on price p2 (cost=0.56..42.63 rows=1 width=27) (actual time=0.005..0.005 rows=0 loops=4891)
Index Cond: ((product_id = p1.product_id) AND ((group_id)::text = ANY ((grps.grp_ids)::text[])) AND (amount = p1.amount) AND (amount = 1))
Filter: ((f_valid = 1) AND ((price < p1.price) OR ((price = p1.price) AND (id < p1.id))))
Rows Removed by Filter: 1
Total runtime: 101.835 ms
(47 rows)
当然，也许 sql 查询不太好，但我想了解为什么 9.4.1 需要更多时间 x4! 你能帮我看看我该怎么办吗？
UPD。我安装了 9.3.6 并尝试了相同的查询。结果比 9.4.1 好得多。
最佳答案
按照@CraigRinger的建议，我将两者放入explain.depesz.com并并排查看。 slow 9.4 query和 fast 9.1 query .
计划基本相同，但有两点很突出。第 4 步(嵌套循环左连接)在 9.4 中花费的时间比在 9.1 中长得多。两者之间似乎有区别，9.4 join 有一个额外的过滤器，Join Filter: ((p1.group_id)::text = ANY ((grps.grp_ids)::text[])).这可能表明两个查询之间存在差异。
然后一切都一样，直到第 17 步，9.4 的一切都变成梨形。它对价格表进行索引扫描需要 273 秒。相反，9.1 执行了一系列更高效的位图扫描。同样，有迹象表明查询可能有所不同。
我愿意...
- 验证您是否运行相同的查询
- 两个版本的表具有相同的索引
- 尤其是价格表
- 运行几次查询以确保您不仅仅对磁盘缓存进行基准测试
如果没有看到查询或架构，我就无能为力。
关于performance - 性能 postgresql 9.1.9/9.3.6 与 9.4.1，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/29059367/