mirror of
https://github.com/apache/impala.git
synced 2026-02-01 12:00:22 -05:00
2040b2621f
HashTable implementation in Impala comprises of contiguous array
of Buckets and each Bucket contains either data or pointer to
linked list of duplicate entries named DuplicateNode.
These are the structures of Bucket and DuplicateNode:
struct DuplicateNode {
bool matched;
DuplicateNode* next;
HtData htdata;
};
struct Bucket {
bool filled;
bool matched;
bool hasDuplicates;
uint32_t hash;
union {
HtData htdata;
DuplicateNode* duplicates;
} bucketData;
};
Size of Bucket is currently 16 bytes and size of DuplicateNode is
24 bytes. If we can remove the booleans from both struct size of
Bucket would reduce to 12 bytes and DuplicateNode will be 16 bytes.
One of the ways we can remove booleans is to fold it into pointers
already part of struct. Pointers store addresses and on
architectures like x86 and ARM the linear address is only 48 bits
long. With level 5 paging Intel is planning to expand it to 57-bit
long which means we can use most significant 7 bits i.e., 58 to 64
bits to store these booleans. This patch reduces the size of Bucket
and DuplicateNode by implementing this folding. However, there is
another requirement regarding Size of Bucket to be power of 2 and
also for the number of buckets in Hash table to be power of 2.
These requirements are for the following reasons:
1. Memory Allocator allocates memory in power of 2 to avoid
internal fragmentation. Hence, num of buckets * sizeof(Buckets)
should be power of 2.
2. Number of buckets being power of 2 enables faster modulo
operation i.e., instead of slow modulo: (hash % N), faster
(hash & (N-1)) can be used.
Due to this, 4 bytes 'hash' field from Bucket is removed and
stored separately in new array hash_array_ in HashTable.
This ensures sizeof(Bucket) is 8 which is power of 2.
New Classes:
------------
As a part of patch, TaggedPointer is introduced which is a template
class to store a pointer and 7-bit tag together in 64 bit integer.
This structure contains the ownership of the pointer and will take care
of allocation and deallocation of the object being pointed to.
However derived classes can opt out of the ownership of the object
and let the client manage it. It's derived classes for Bucket and
DuplicateNode do the same. These classes are TaggedBucketData and
TaggedDuplicateNode.
Benchmark:
----------
As a part of this patch a new Micro Benchmark for HashTable has
been introduced, which will help in measuring these:
1. Runtime for building hash table and probing it.
2. Memory consumed after building the Table.
This would help measuring the impact of changes to the HashTable's
data structure and algorithm.
Saw 25-30% reduction in memory consumed and no significant
difference in performance (0.91X-1.2X).
Other Benchmarks:
1. Billion row Synthetic benchmark on single node, single daemon:
a. 2-3% improvement in Join GEOMEAN for Probe benchmark.
b. 17% and 21% reduction in PeakMemoryUsage and
CumulativeBytes allocated respectively
2. TPCH-42: 0-1.5% improvement in GEOMEAN runtime
Change-Id: I72912ae9353b0d567a976ca712d2d193e035df9b
Reviewed-on: http://gerrit.cloudera.org:8080/17592
Reviewed-by: Zoltan Borok-Nagy <boroknagyz@cloudera.com>
Tested-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>
128 lines
5.7 KiB
Plaintext
128 lines
5.7 KiB
Plaintext
====
|
|
---- QUERY
|
|
# Spilling broadcast join between lineitem and orders.
|
|
# Simplified from TPC-Q21.
|
|
set buffer_pool_limit=50m;
|
|
select straight_join s_name, count(*)
|
|
from lineitem join /*+broadcast*/ orders on o_orderkey = l_orderkey
|
|
join supplier on s_suppkey = l_suppkey
|
|
where o_orderstatus = 'F'
|
|
group by s_name
|
|
order by count(*) desc, s_name
|
|
limit 20
|
|
---- RESULTS
|
|
'Supplier#000002293',350
|
|
'Supplier#000006337',349
|
|
'Supplier#000008446',349
|
|
'Supplier#000009636',349
|
|
'Supplier#000006566',347
|
|
'Supplier#000000219',346
|
|
'Supplier#000000870',346
|
|
'Supplier#000001689',346
|
|
'Supplier#000002493',345
|
|
'Supplier#000000208',344
|
|
'Supplier#000008565',343
|
|
'Supplier#000001941',342
|
|
'Supplier#000006467',341
|
|
'Supplier#000007489',341
|
|
'Supplier#000007079',340
|
|
'Supplier#000000326',339
|
|
'Supplier#000003041',339
|
|
'Supplier#000002196',338
|
|
'Supplier#000005229',338
|
|
'Supplier#000006159',338
|
|
---- TYPES
|
|
STRING,BIGINT
|
|
---- RUNTIME_PROFILE
|
|
# Verify that at least one of the joins was spilled.
|
|
row_regex: .*SpilledPartitions: .* \([1-9][0-9]*\)
|
|
====
|
|
---- QUERY
|
|
# Spilling broadcast join that will repartition.
|
|
# Force a bushy plan where join of customer and orders feeds into the broadcast join.
|
|
set buffer_pool_limit=59m;
|
|
select straight_join o_orderkey, o_custkey, o_orderstatus, o_totalprice, o_orderdate,
|
|
o_orderpriority, o_clerk, o_shippriority, c_name, c_nationkey, c_address, c_comment
|
|
from lineitem join /*+broadcast*/
|
|
(select *
|
|
from orders
|
|
join customer on o_custkey = c_custkey) v on o_orderkey = l_orderkey
|
|
group by 1,2,3,4,5,6,7,8,9,10,11,12
|
|
order by count(*) desc, o_orderkey
|
|
limit 10
|
|
---- RESULTS
|
|
7,39136,'O',252004.18,'1996-01-10','2-HIGH','Clerk#000000470',0,'Customer#000039136',5,'afZJC1mWpwvsfKT0211ZD6NQXVGETfl','y? express theodolites haggle against the bold instructions. slyly regular accoun'
|
|
68,28547,'O',330793.52,'1998-04-18','3-MEDIUM','Clerk#000000440',0,'Customer#000028547',1,'AeWmD3BLrsSkmRY7O,wbB75i6Ll','y regular foxes nag quickly after the express, regular deposits. regular, bold requests nod furi'
|
|
129,71134,'F',261013.14,'1992-11-19','5-LOW','Clerk#000000859',0,'Customer#000071134',8,'QrIUhKsAaaU8pLxHNHpG mN1F n0eWQxE2',' ideas. furiously pending packages dazzle blithely. special platelets nod f'
|
|
164,779,'F',301925.76,'1992-10-21','5-LOW','Clerk#000000209',0,'Customer#000000779',5,'2cTZiS4ulZ74edT,RmDnh4ZaCrphMMh Ff2','old dependencies. pains haggle fluffily carefull'
|
|
194,61724,'F',176707.84,'1992-04-05','3-MEDIUM','Clerk#000000352',0,'Customer#000061724',8,'konflahJ4RK4riiSPR3 duNNLr5n Es7gF','regular accounts. carefully ironic requests '
|
|
225,33031,'P',220441.09,'1995-05-25','1-URGENT','Clerk#000000177',0,'Customer#000033031',8,'YFlDbKd1jeGbXz WgshqgAslqv,T4Rfe','e quickly. slyly special pinto beans nag quickly? carefully ironic de'
|
|
226,127466,'F',313134.62,'1993-03-10','2-HIGH','Clerk#000000756',0,'Customer#000127466',17,'0,cSa,GNNBHU','tes haggle furiously during the unusual, even'
|
|
322,133546,'F',205623.50,'1992-03-19','1-URGENT','Clerk#000000158',0,'Customer#000133546',5,'J0vOJt9OkiZ0R3kE,RtYvv','. final gifts are slyly pending packages. even instructio'
|
|
326,75986,'O',327413.14,'1995-06-04','2-HIGH','Clerk#000000466',0,'Customer#000075986',21,'h3vC,A ibqYy2VagkCNfRiQ','y after the carefully bold instructions. special, regular instructions cajo'
|
|
354,138268,'O',217160.72,'1996-03-14','2-HIGH','Clerk#000000511',0,'Customer#000138268',9,'HCMHttdHqpeDYf','inal, express deposits against the slowly'
|
|
---- TYPES
|
|
BIGINT, BIGINT, STRING, DECIMAL, STRING, STRING, STRING, INT, STRING, SMALLINT, STRING, STRING
|
|
---- RUNTIME_PROFILE
|
|
# Verify that at least one of the joins was spilled and repartitioned.
|
|
row_regex: .*SpilledPartitions: .* \([1-9][0-9]*\)
|
|
row_regex: .*NumRepartitions: .* \([1-9][0-9]*\)
|
|
====
|
|
---- QUERY
|
|
# Spilling broadcast join with empty probe-side partitions.
|
|
set buffer_pool_limit=46m;
|
|
select straight_join count(*)
|
|
from
|
|
lineitem a join /*+broadcast*/ lineitem b on a.l_orderkey = b.l_orderkey
|
|
where
|
|
a.l_partkey = 1 and b.l_orderkey < 1000000;
|
|
---- RESULTS
|
|
19
|
|
---- TYPES
|
|
BIGINT
|
|
---- RUNTIME_PROFILE
|
|
# Verify that at least one of the joins was spilled and that same hash table builds
|
|
# were *not* skipped - that optimization is disabled for shared broadcast joins.
|
|
row_regex: .*SpilledPartitions: .* \([1-9][0-9]*\)
|
|
row_regex: .*NumHashTableBuildsSkipped: 0
|
|
====
|
|
---- QUERY
|
|
# Spilling broadcast join that has some empty build partitions.
|
|
# This needs to be calibrated so that the join has enough memory to
|
|
# fit all of the duplicate values for a given key in memory.
|
|
set buffer_pool_limit=60m;
|
|
select straight_join s_name, o_clerk, count(*)
|
|
from lineitem
|
|
join /*+broadcast*/ orders on (o_orderkey % 8) = l_orderkey
|
|
join supplier on s_suppkey = l_suppkey
|
|
where o_orderstatus = 'F'
|
|
group by s_name, o_clerk
|
|
order by count(*) desc, s_name, o_clerk limit 20;
|
|
---- RESULTS
|
|
'Supplier#000000035','Clerk#000000337',126
|
|
'Supplier#000001759','Clerk#000000225',126
|
|
'Supplier#000002269','Clerk#000000225',126
|
|
'Supplier#000003074','Clerk#000000225',126
|
|
'Supplier#000003928','Clerk#000000337',126
|
|
'Supplier#000007758','Clerk#000000225',126
|
|
'Supplier#000008571','Clerk#000000337',126
|
|
'Supplier#000009440','Clerk#000000225',126
|
|
'Supplier#000009607','Clerk#000000225',126
|
|
'Supplier#000009799','Clerk#000000225',126
|
|
'Supplier#000000638','Clerk#000000461',125
|
|
'Supplier#000000650','Clerk#000000718',125
|
|
'Supplier#000001534','Clerk#000000461',125
|
|
'Supplier#000001798','Clerk#000000718',125
|
|
'Supplier#000001883','Clerk#000000718',125
|
|
'Supplier#000003474','Clerk#000000718',125
|
|
'Supplier#000003701','Clerk#000000461',125
|
|
'Supplier#000004633','Clerk#000000461',125
|
|
'Supplier#000006540','Clerk#000000718',125
|
|
'Supplier#000007311','Clerk#000000461',125
|
|
---- TYPES
|
|
STRING, STRING, BIGINT
|
|
---- RUNTIME_PROFILE
|
|
# Verify that at least one of the joins was spilled.
|
|
row_regex: .*SpilledPartitions: .* \([1-9][0-9]*\)
|
|
====
|