CPU modern duwe akeh inti. Wis pirang-pirang taun, aplikasi wis ngirim pitakon menyang database kanthi paralel. Yen iku pitakonan laporan ing sawetara larik ing meja, mlaku luwih cepet nalika nggunakake sawetara CPU, lan PostgreSQL wis bisa nindakake iki wiwit versi 9.6.
Butuh 3 taun kanggo ngleksanakake fitur query paralel - kita kudu nulis ulang kode ing macem-macem tahapan eksekusi query. PostgreSQL 9.6 ngenalake infrastruktur kanggo nambah kode kasebut. Ing versi sabanjure, jinis pitakon liyane dieksekusi kanthi paralel.
Watesan
- Aja ngaktifake eksekusi paralel yen kabeh inti wis sibuk, yen panjaluk liyane bakal alon.
- Sing paling penting, pangolahan paralel kanthi nilai WORK_MEM sing dhuwur nggunakake akeh memori - saben gabung utawa ngurutake nggunakake memori work_mem.
- Pitakonan OLTP latency kurang ora bisa dicepetake kanthi eksekusi paralel. Lan yen pitakon ngasilake siji baris, pangolahan paralel mung bakal alon.
- Pangembang seneng nggunakake pathokan TPC-H. Mungkin sampeyan duwe pitakon sing padha kanggo eksekusi paralel sing sampurna.
- Mung pitakon PILIH tanpa ngunci predikat sing dieksekusi kanthi paralel.
- Kadhangkala indeksasi sing tepat luwih apik tinimbang mindhai tabel urutan ing mode paralel.
- Ngaso pitakon lan kursor ora didhukung.
- Fungsi jendhela lan fungsi agregat sing diurutake ora sejajar.
- Sampeyan ora entuk apa-apa ing beban kerja I / O.
- Ora ana algoritma ngurutake paralel. Nanging pitakon kanthi macem-macem bisa dieksekusi kanthi paralel ing sawetara aspek.
- Ganti CTE (WITH ...) karo SELECT nested kanggo ngaktifake pangolahan paralel.
- Pembungkus data pihak katelu durung ndhukung pangolahan paralel (nanging bisa!)
- FULL OUTER JOIN ora didhukung.
- max_rows mateni pangolahan paralel.
- Yen pitakonan nduweni fungsi sing ora ditandhani PARALLEL SAFE, iku bakal dadi siji Utas.
- Tingkat isolasi transaksi SERIALIZABLE mateni pangolahan paralel.
Lingkungan tes
Pangembang PostgreSQL nyoba nyuda wektu nanggepi pitakon benchmark TPC-H. Download pathokan lan . Iki minangka panggunaan ora resmi saka pathokan TPC-H - ora kanggo basis data utawa perbandingan hardware.
- Download TPC-H_Tools_v2.17.3.zip (utawa versi anyar) .
- Ganti jeneng makefile.suite dadi Makefile lan ganti kaya sing diterangake ing kene: . Kompilasi kode kanthi printah make.
- Nggawe data:
./dbgen -s 10nggawe database 23 GB. Iki cukup kanggo ndeleng prabédan ing kinerja pitakon paralel lan non-paralel. - Ngonversi file
tblвcsv с forиsed. - Kloning repositori
pg_tpchlan nyalin filecsvвpg_tpch/dss/data. - Gawe pitakon kanthi prentah
qgen. - Muat data menyang database kanthi printah
./tpch.sh.
Pemindaian urutan paralel
Bisa uga luwih cepet ora amarga maca paralel, nanging amarga data nyebar ing akeh inti CPU. Ing sistem operasi modern, file data PostgreSQL disimpen kanthi apik. Kanthi maca sadurunge, bisa entuk blok sing luwih gedhe saka panyimpenan tinimbang panjaluk daemon PG. Mulane, kinerja query ora diwatesi dening disk I/O. Iku nganggo siklus CPU kanggo:
- maca larik siji-siji saka kaca tabel;
- mbandhingake nilai lan kahanan string
WHERE.
Ayo mbukak pitakonan prasaja select:
tpch=# explain analyze select l_quantity as sum_qty from lineitem where l_shipdate <= date '1998-12-01' - interval '105' day;
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------
Seq Scan on lineitem (cost=0.00..1964772.00 rows=58856235 width=5) (actual time=0.014..16951.669 rows=58839715 loops=1)
Filter: (l_shipdate <= '1998-08-18 00:00:00'::timestamp without time zone)
Rows Removed by Filter: 1146337
Planning Time: 0.203 ms
Execution Time: 19035.100 msPindai sekuensial ngasilake akeh banget larik tanpa agregasi, mula pitakon dieksekusi dening inti CPU siji.
Yen sampeyan nambah SUM(), sampeyan bisa ndeleng manawa rong alur kerja bakal mbantu nyepetake pitakon:
explain analyze select sum(l_quantity) as sum_qty from lineitem where l_shipdate <= date '1998-12-01' - interval '105' day;
QUERY PLAN
----------------------------------------------------------------------------------------------------------------------------------------------------
Finalize Aggregate (cost=1589702.14..1589702.15 rows=1 width=32) (actual time=8553.365..8553.365 rows=1 loops=1)
-> Gather (cost=1589701.91..1589702.12 rows=2 width=32) (actual time=8553.241..8555.067 rows=3 loops=1)
Workers Planned: 2
Workers Launched: 2
-> Partial Aggregate (cost=1588701.91..1588701.92 rows=1 width=32) (actual time=8547.546..8547.546 rows=1 loops=3)
-> Parallel Seq Scan on lineitem (cost=0.00..1527393.33 rows=24523431 width=5) (actual time=0.038..5998.417 rows=19613238 loops=3)
Filter: (l_shipdate <= '1998-08-18 00:00:00'::timestamp without time zone)
Rows Removed by Filter: 382112
Planning Time: 0.241 ms
Execution Time: 8555.131 msAgregasi paralel
Node Parallel Seq Scan ngasilake baris kanggo agregasi parsial. Node "Partial Aggregate" trims garis iki nggunakake SUM(). Ing pungkasan, counter SUM saka saben proses buruh diklumpukake dening simpul "Kumpul".
Asil pungkasan diitung kanthi simpul "Rampungake Agregat". Yen sampeyan duwe fungsi agregasi dhewe, aja lali tandhani minangka "aman paralel".
Jumlah pangolahan buruh
Jumlah pangolahan buruh bisa ditambah tanpa miwiti maneh server:
explain analyze select sum(l_quantity) as sum_qty from lineitem where l_shipdate <= date '1998-12-01' - interval '105' day;
QUERY PLAN
----------------------------------------------------------------------------------------------------------------------------------------------------
Finalize Aggregate (cost=1589702.14..1589702.15 rows=1 width=32) (actual time=8553.365..8553.365 rows=1 loops=1)
-> Gather (cost=1589701.91..1589702.12 rows=2 width=32) (actual time=8553.241..8555.067 rows=3 loops=1)
Workers Planned: 2
Workers Launched: 2
-> Partial Aggregate (cost=1588701.91..1588701.92 rows=1 width=32) (actual time=8547.546..8547.546 rows=1 loops=3)
-> Parallel Seq Scan on lineitem (cost=0.00..1527393.33 rows=24523431 width=5) (actual time=0.038..5998.417 rows=19613238 loops=3)
Filter: (l_shipdate <= '1998-08-18 00:00:00'::timestamp without time zone)
Rows Removed by Filter: 382112
Planning Time: 0.241 ms
Execution Time: 8555.131 msAna apa ing kene? Ana 2 kaping pangolahan karya, lan panjalukan dadi mung 1,6599 kaping luwih cepet. Petungan sing menarik. Kita duwe 2 pangolahan buruh lan 1 pimpinan. Sawise owah-owahan dadi 4 + 1.
kacepetan maksimum kita saka Processing podo: 5/3 = 1,66 (6) kaping.
Carane ora iku bisa?
Pangolahan
Panyuwunan eksekusi mesthi diwiwiti kanthi proses utama. Pimpinan nindakake kabeh sing ora paralel lan sawetara pangolahan paralel. Proses liyane sing nindakake panjalukan sing padha diarani proses buruh. Pangolahan paralel nggunakake infrastruktur (saka versi 9.4). Amarga bagean PostgreSQL liyane nggunakake proses tinimbang benang, pitakon kanthi 3 proses buruh bisa 4 kaping luwih cepet tinimbang pangolahan tradisional.
Interaksi
Proses buruh komunikasi karo pimpinan liwat antrian pesen (adhedhasar memori sing dienggo bareng). Saben proses duwe 2 antrian: kanggo kesalahan lan tuple.
Pira alur kerja sing dibutuhake?
Watesan minimal ditemtokake dening parameter . Panyuwunan runner banjur njupuk pangolahan buruh saka blumbang diwatesi dening parameter . Watesan pungkasan yaiku , yaiku, jumlah total pangolahan latar mburi.
Yen ora bisa menehi proses buruh, pangolahan bakal dadi siji-proses.
Perencana pitakon bisa nyuda alur kerja gumantung saka ukuran tabel utawa indeks. Ana paramèter kanggo iki и .
set min_parallel_table_scan_size='8MB'
8MB table => 1 worker
24MB table => 2 workers
72MB table => 3 workers
x => log(x / min_parallel_table_scan_size) / log(3) + 1 workerSaben wektu meja punika 3 kaping luwih gedhe saka min_parallel_(index|table)_scan_size, Postgres nambahake proses buruh. Jumlah alur kerja ora adhedhasar biaya. Ketergantungan bunder nggawe implementasine rumit. Nanging, planner nggunakake aturan prasaja.
Ing laku, aturan kasebut ora mesthi cocok kanggo produksi, supaya sampeyan bisa ngganti jumlah pangolahan buruh kanggo tabel tartamtu: ALTER TABLE ... SET (parallel_workers = N).
Napa pangolahan paralel ora digunakake?
Saliyane dhaptar watesan sing dawa, ana uga pamriksa biaya:
- kanggo ngindhari pangolahan paralel saka panjalukan singkat. Parameter iki ngira wektu kanggo nyiapake memori, miwiti proses, lan ijol-ijolan data awal.
: komunikasi antarane pimpinan lan buruh bisa ditundha ing babagan jumlah tuple saka proses karya. Parameter iki ngitung biaya ijol-ijolan data.
Nested Loop Gabung
PostgreSQL 9.6+ может выполнять вложенные циклы параллельно — это простая операция.
explain (costs off) select c_custkey, count(o_orderkey)
from customer left outer join orders on
c_custkey = o_custkey and o_comment not like '%special%deposits%'
group by c_custkey;
QUERY PLAN
--------------------------------------------------------------------------------------
Finalize GroupAggregate
Group Key: customer.c_custkey
-> Gather Merge
Workers Planned: 4
-> Partial GroupAggregate
Group Key: customer.c_custkey
-> Nested Loop Left Join
-> Parallel Index Only Scan using customer_pkey on customer
-> Index Scan using idx_orders_custkey on orders
Index Cond: (customer.c_custkey = o_custkey)
Filter: ((o_comment)::text !~~ '%special%deposits%'::text)Koleksi kasebut dumadi ing tahap pungkasan, mula Nested Loop Left Join minangka operasi paralel. Indeks Paralel Mung Pindai dikenalake mung ing versi 10. Kerjane padha karo pemindaian serial paralel. kahanan c_custkey = o_custkey maca siji pesenan saben senar klien. Dadi ora sejajar.
Hash Gabung
Saben proses buruh nggawe tabel hash dhewe nganti PostgreSQL 11. Lan yen ana luwih saka papat proses kasebut, kinerja ora bakal nambah. Ing versi anyar, tabel hash dienggo bareng. Saben proses buruh bisa nggunakake WORK_MEM kanggo nggawe tabel hash.
select
l_shipmode,
sum(case
when o_orderpriority = '1-URGENT'
or o_orderpriority = '2-HIGH'
then 1
else 0
end) as high_line_count,
sum(case
when o_orderpriority <> '1-URGENT'
and o_orderpriority <> '2-HIGH'
then 1
else 0
end) as low_line_count
from
orders,
lineitem
where
o_orderkey = l_orderkey
and l_shipmode in ('MAIL', 'AIR')
and l_commitdate < l_receiptdate
and l_shipdate < l_commitdate
and l_receiptdate >= date '1996-01-01'
and l_receiptdate < date '1996-01-01' + interval '1' year
group by
l_shipmode
order by
l_shipmode
LIMIT 1;
QUERY PLAN
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Limit (cost=1964755.66..1964961.44 rows=1 width=27) (actual time=7579.592..7922.997 rows=1 loops=1)
-> Finalize GroupAggregate (cost=1964755.66..1966196.11 rows=7 width=27) (actual time=7579.590..7579.591 rows=1 loops=1)
Group Key: lineitem.l_shipmode
-> Gather Merge (cost=1964755.66..1966195.83 rows=28 width=27) (actual time=7559.593..7922.319 rows=6 loops=1)
Workers Planned: 4
Workers Launched: 4
-> Partial GroupAggregate (cost=1963755.61..1965192.44 rows=7 width=27) (actual time=7548.103..7564.592 rows=2 loops=5)
Group Key: lineitem.l_shipmode
-> Sort (cost=1963755.61..1963935.20 rows=71838 width=27) (actual time=7530.280..7539.688 rows=62519 loops=5)
Sort Key: lineitem.l_shipmode
Sort Method: external merge Disk: 2304kB
Worker 0: Sort Method: external merge Disk: 2064kB
Worker 1: Sort Method: external merge Disk: 2384kB
Worker 2: Sort Method: external merge Disk: 2264kB
Worker 3: Sort Method: external merge Disk: 2336kB
-> Parallel Hash Join (cost=382571.01..1957960.99 rows=71838 width=27) (actual time=7036.917..7499.692 rows=62519 loops=5)
Hash Cond: (lineitem.l_orderkey = orders.o_orderkey)
-> Parallel Seq Scan on lineitem (cost=0.00..1552386.40 rows=71838 width=19) (actual time=0.583..4901.063 rows=62519 loops=5)
Filter: ((l_shipmode = ANY ('{MAIL,AIR}'::bpchar[])) AND (l_commitdate < l_receiptdate) AND (l_shipdate < l_commitdate) AND (l_receiptdate >= '1996-01-01'::date) AND (l_receiptdate < '1997-01-01 00:00:00'::timestamp without time zone))
Rows Removed by Filter: 11934691
-> Parallel Hash (cost=313722.45..313722.45 rows=3750045 width=20) (actual time=2011.518..2011.518 rows=3000000 loops=5)
Buckets: 65536 Batches: 256 Memory Usage: 3840kB
-> Parallel Seq Scan on orders (cost=0.00..313722.45 rows=3750045 width=20) (actual time=0.029..995.948 rows=3000000 loops=5)
Planning Time: 0.977 ms
Execution Time: 7923.770 msPitakonan 12 saka TPC-H kanthi jelas nuduhake sambungan hash paralel. Saben proses buruh nyumbang kanggo nggawe tabel hash umum.
Gabung Gabung
Gabungan gabungan ora sejajar. Aja kuwatir yen iki langkah pungkasan saka pitakonan - isih bisa mlaku bebarengan.
-- Query 2 from TPC-H
explain (costs off) select s_acctbal, s_name, n_name, p_partkey, p_mfgr, s_address, s_phone, s_comment
from part, supplier, partsupp, nation, region
where
p_partkey = ps_partkey
and s_suppkey = ps_suppkey
and p_size = 36
and p_type like '%BRASS'
and s_nationkey = n_nationkey
and n_regionkey = r_regionkey
and r_name = 'AMERICA'
and ps_supplycost = (
select
min(ps_supplycost)
from partsupp, supplier, nation, region
where
p_partkey = ps_partkey
and s_suppkey = ps_suppkey
and s_nationkey = n_nationkey
and n_regionkey = r_regionkey
and r_name = 'AMERICA'
)
order by s_acctbal desc, n_name, s_name, p_partkey
LIMIT 100;
QUERY PLAN
----------------------------------------------------------------------------------------------------------
Limit
-> Sort
Sort Key: supplier.s_acctbal DESC, nation.n_name, supplier.s_name, part.p_partkey
-> Merge Join
Merge Cond: (part.p_partkey = partsupp.ps_partkey)
Join Filter: (partsupp.ps_supplycost = (SubPlan 1))
-> Gather Merge
Workers Planned: 4
-> Parallel Index Scan using <strong>part_pkey</strong> on part
Filter: (((p_type)::text ~~ '%BRASS'::text) AND (p_size = 36))
-> Materialize
-> Sort
Sort Key: partsupp.ps_partkey
-> Nested Loop
-> Nested Loop
Join Filter: (nation.n_regionkey = region.r_regionkey)
-> Seq Scan on region
Filter: (r_name = 'AMERICA'::bpchar)
-> Hash Join
Hash Cond: (supplier.s_nationkey = nation.n_nationkey)
-> Seq Scan on supplier
-> Hash
-> Seq Scan on nation
-> Index Scan using idx_partsupp_suppkey on partsupp
Index Cond: (ps_suppkey = supplier.s_suppkey)
SubPlan 1
-> Aggregate
-> Nested Loop
Join Filter: (nation_1.n_regionkey = region_1.r_regionkey)
-> Seq Scan on region region_1
Filter: (r_name = 'AMERICA'::bpchar)
-> Nested Loop
-> Nested Loop
-> Index Scan using idx_partsupp_partkey on partsupp partsupp_1
Index Cond: (part.p_partkey = ps_partkey)
-> Index Scan using supplier_pkey on supplier supplier_1
Index Cond: (s_suppkey = partsupp_1.ps_suppkey)
-> Index Scan using nation_pkey on nation nation_1
Index Cond: (n_nationkey = supplier_1.s_nationkey)Node "Gabung Gabung" dumunung ing ndhuwur "Gather Merge". Dadi gabung ora nggunakake pangolahan paralel. Nanging simpul "Indeks Paralel" isih mbantu babagan segmen kasebut part_pkey.
Sambungan dening bagean
Ing PostgreSQL 11 dipatèni minangka standar: wis jadwal larang banget. Tabel kanthi partisi sing padha bisa digabung karo partisi kanthi partisi. Kanthi cara iki Postgres bakal nggunakake tabel hash sing luwih cilik. Saben sambungan bagean bisa podo karo.
tpch=# set enable_partitionwise_join=t;
tpch=# explain (costs off) select * from prt1 t1, prt2 t2
where t1.a = t2.b and t1.b = 0 and t2.b between 0 and 10000;
QUERY PLAN
---------------------------------------------------
Append
-> Hash Join
Hash Cond: (t2.b = t1.a)
-> Seq Scan on prt2_p1 t2
Filter: ((b >= 0) AND (b <= 10000))
-> Hash
-> Seq Scan on prt1_p1 t1
Filter: (b = 0)
-> Hash Join
Hash Cond: (t2_1.b = t1_1.a)
-> Seq Scan on prt2_p2 t2_1
Filter: ((b >= 0) AND (b <= 10000))
-> Hash
-> Seq Scan on prt1_p2 t1_1
Filter: (b = 0)
tpch=# set parallel_setup_cost = 1;
tpch=# set parallel_tuple_cost = 0.01;
tpch=# explain (costs off) select * from prt1 t1, prt2 t2
where t1.a = t2.b and t1.b = 0 and t2.b between 0 and 10000;
QUERY PLAN
-----------------------------------------------------------
Gather
Workers Planned: 4
-> Parallel Append
-> Parallel Hash Join
Hash Cond: (t2_1.b = t1_1.a)
-> Parallel Seq Scan on prt2_p2 t2_1
Filter: ((b >= 0) AND (b <= 10000))
-> Parallel Hash
-> Parallel Seq Scan on prt1_p2 t1_1
Filter: (b = 0)
-> Parallel Hash Join
Hash Cond: (t2.b = t1.a)
-> Parallel Seq Scan on prt2_p1 t2
Filter: ((b >= 0) AND (b <= 10000))
-> Parallel Hash
-> Parallel Seq Scan on prt1_p1 t1
Filter: (b = 0)Ingkang utama yaiku sambungan ing bagean kasebut sejajar mung yen bagean kasebut cukup gedhe.
Lampiran Paralel
bisa digunakake tinimbang pamblokiran beda ing workflows beda. Iki biasane kedadeyan karo pitakon UNION ALL. Kerugian kurang paralelisme, amarga saben proses buruh mung ngolah 1 panjaluk.
Ana 2 pangolahan buruh sing mlaku ing kene, sanajan 4 diaktifake.
tpch=# explain (costs off) select sum(l_quantity) as sum_qty from lineitem where l_shipdate <= date '1998-12-01' - interval '105' day union all select sum(l_quantity) as sum_qty from lineitem where l_shipdate <= date '2000-12-01' - interval '105' day;
QUERY PLAN
------------------------------------------------------------------------------------------------
Gather
Workers Planned: 2
-> Parallel Append
-> Aggregate
-> Seq Scan on lineitem
Filter: (l_shipdate <= '2000-08-18 00:00:00'::timestamp without time zone)
-> Aggregate
-> Seq Scan on lineitem lineitem_1
Filter: (l_shipdate <= '1998-08-18 00:00:00'::timestamp without time zone)Variabel sing paling penting
- WORK_MEM mbatesi memori saben proses, ora mung pitakon: work_mem pangolahan sambungan = akeh memori.
- - pirang-pirang buruh ngolah program sing bakal digunakake kanggo pangolahan paralel saka rencana kasebut.
- - nyetel jumlah total pangolahan buruh kanggo jumlah inti CPU ing server.
- - padha, nanging kanggo pangolahan karya podo.
Hasil
Ing versi 9.6, pangolahan paralel bisa ningkatake kinerja pitakon kompleks sing mindai akeh larik utawa indeks. Ing PostgreSQL 10, pangolahan paralel diaktifake kanthi gawan. Elinga mateni ing server kanthi beban kerja OLTP sing gedhe. Pindai urut utawa pindai indeks nggunakake akeh sumber daya. Yen sampeyan ora mbukak laporan babagan kabeh set data, sampeyan bisa nambah kinerja pitakon kanthi mung nambah indeks sing ilang utawa nggunakake partisi sing tepat.
referensi
Source: www.habr.com