Support Questions
Find answers, ask questions, and share your expertise
Announcements
Alert: Welcome to the Unified Cloudera Community. Former HCC members be sure to read and learn how to activate your account here.

Imapla query is slow, where is the bottleneck?

Solved Go to solution
Highlighted

Imapla query is slow, where is the bottleneck?

Explorer

I have this query:

SELECT *
FROM ft_test, ft_wafer
WHERE ft_test_parquet.id = ft_wafer_parquet.id
and month = 1
and day = 8 
and param = 2913;

It took about 3.5s. I want this to be faster. I have ready used techniques like Parquet, partition, compute stats

 

Here is the profile. Could someone look through this and find the bottleneck? Thanks.

 

Query Runtime Profile:
Query (id=dc4c084615fbf9bb:4261466f00000000):
  Summary:
    Session ID: 5d4edbf63653cdf6:1a59ff5354c9d4bd
    Session Type: BEESWAX
    Start Time: 2017-05-25 16:31:25.121391000
    End Time: 2017-05-25 16:31:28.584404000
    Query Type: QUERY
    Query State: FINISHED
    Query Status: OK
    Impala Version: impalad version 2.7.0 RELEASE (build a535b583202c4a81080098a10f952d377af1949d)
    User: root
    Connected User: root
    Delegated User:
    Network Address: ::ffff:127.0.0.1:58546
    Default Db: inspex
    Sql Statement: select *
FROM ft_test_partition, ft_wafer_parquet
WHERE ft_test_partition.file = ft_wafer_parquet.file
and month = 1
and day = 8
and param = 2913 limit 100
    Coordinator: mapr1:22000
    Query Options (non default):
    Plan:
----------------
Estimated Per-Host Requirements: Memory=704.67MB VCores=2

04:EXCHANGE [UNPARTITIONED]
|  limit: 100
|  hosts=1 per-host-mem=unavailable
|  tuple-ids=1,0 row-size=800B cardinality=1
|
02:HASH JOIN [INNER JOIN, BROADCAST]
|  hash predicates: ft_wafer_parquet.file = ft_test_partition.file
|  runtime filters: RF000 <- ft_test_partition.file
|  limit: 100
|  hosts=1 per-host-mem=690.00KB
|  tuple-ids=1,0 row-size=800B cardinality=1
|
|--03:EXCHANGE [BROADCAST]
|  |  hosts=1 per-host-mem=0B
|  |  tuple-ids=0 row-size=78B cardinality=8235
|  |
|  00:SCAN HDFS [inspex.ft_test_partition, RANDOM]
|     partitions=1/29 files=1 size=171.69MB
|     predicates: param = 2913
|     table stats: 813365826 rows total
|     column stats: all
|     hosts=1 per-host-mem=704.00MB
|     tuple-ids=0 row-size=78B cardinality=8235
|
01:SCAN HDFS [inspex.ft_wafer_parquet, RANDOM]
   partitions=1/1 files=1 size=66.83KB
   runtime filters: RF000 -> ft_wafer_parquet.file
   table stats: 1500 rows total
   column stats: all
   hosts=1 per-host-mem=192.00MB
   tuple-ids=1 row-size=722B cardinality=1500
----------------
    Estimated Per-Host Mem: 738904067
    Estimated Per-Host VCores: 2
    Request Pool: default-pool
    Admission result: Admitted immediately
    ExecSummary:
Operator          #Hosts   Avg Time   Max Time  #Rows  Est. #Rows   Peak Mem  Est. Peak Mem  Detail
----------------------------------------------------------------------------------------------------------------------
04:EXCHANGE            1    0.000ns    0.000ns    100           1          0        -1.00 B  UNPARTITIONED
02:HASH JOIN           1   42.999ms   42.999ms      0           1    3.29 MB      690.00 KB  INNER JOIN, BROADCAST
|--03:EXCHANGE         1  999.990us  999.990us  9.15K       8.23K          0              0  BROADCAST
|  00:SCAN HDFS        1    2s973ms    2s973ms  9.15K       8.23K  106.05 MB      704.00 MB  inspex.ft_test_partition
01:SCAN HDFS           1   16.999ms   16.999ms  1.02K       1.50K    1.78 MB      192.00 MB  inspex.ft_wafer_parquet
    Planner Timeline: 15.315ms
       - Analysis finished: 5.081ms (5.081ms)
       - Equivalence classes computed: 5.601ms (519.374us)
       - Single node plan created: 9.054ms (3.453ms)
       - Runtime filters computed: 9.409ms (354.377us)
       - Distributed plan created: 11.507ms (2.098ms)
       - Planning finished: 15.315ms (3.808ms)
    Query Timeline: 3s463ms
       - Start execution: 0.000ns (0.000ns)
       - Planning finished: 17.999ms (17.999ms)
       - Submit for admission: 17.999ms (0.000ns)
       - Completed admission: 17.999ms (0.000ns)
       - Ready to start 2 remote fragments: 18.999ms (999.990us)
       - All 2 remote fragments started: 19.999ms (999.990us)
       - Rows available: 3s246ms (3s226ms)
       - First row fetched: 3s346ms (99.999ms)
       - Unregister query: 3s462ms (115.998ms)
     - ComputeScanRangeAssignmentTimer: 0.000ns
  ImpalaServer:
     - ClientFetchWaitTimer: 214.997ms
     - RowMaterializationTimer: 999.990us
  Execution Profile dc4c084615fbf9bb:4261466f00000000:(Total: 3s228ms, non-child: 0.000ns, % non-child: 0.00%)
    Number of filters: 1
    Filter routing table:
 ID  Src. Node  Tgt. Node(s)  Targets  Target type  Partition filter  Pending (Expected)  First arrived  Completed   Enabled
----------------------------------------------------------------------------------------------------------------------------
  0          2             1        1        LOCAL             false               0 (1)            N/A        N/A      true
    Fragment start latencies: Count: 2, 25th %-ile: 0, 50th %-ile: 0, 75th %-ile: 1ms, 90th %-ile: 1ms, 95th %-ile: 1ms, 99.9th %-ile: 1ms
    Final filter table:
 ID  Src. Node  Tgt. Node(s)  Targets  Target type  Partition filter  Pending (Expected)  First arrived  Completed   Enabled
----------------------------------------------------------------------------------------------------------------------------
  0          2             1        1        LOCAL             false               0 (1)            N/A        N/A      true
    Per Node Peak Memory Usage: mapr1:22000(108.65 MB)
     - FiltersReceived: 0 (0)
     - FinalizationTimer: 0.000ns
    Coordinator Fragment F02:(Total: 3s226ms, non-child: 0.000ns, % non-child: 0.00%)
      MemoryUsage(500.000ms): 16.00 KB, 16.00 KB, 16.00 KB, 16.00 KB, 16.00 KB, 16.00 KB, 16.00 KB
       - AverageThreadTokens: 0.00
       - BloomFilterBytes: 0
       - PeakMemoryUsage: 209.83 KB (214864)
       - PerHostPeakMemUsage: 0
       - PrepareTime: 0.000ns
       - RowsProduced: 0 (0)
       - TotalCpuTime: 101.999ms
       - TotalNetworkReceiveTime: 3s226ms
       - TotalNetworkSendTime: 0.000ns
       - TotalStorageWaitTime: 0.000ns
      BlockMgr:
         - BlockWritesOutstanding: 0 (0)
         - BlocksCreated: 48 (48)
         - BlocksRecycled: 0 (0)
         - BufferedPins: 0 (0)
         - BytesWritten: 0
         - MaxBlockSize: 8.00 MB (8388608)
         - MemoryLimit: 12.21 GB (13111148544)
         - PeakMemoryUsage: 256.00 KB (262144)
         - TotalBufferWaitTime: 0.000ns
         - TotalEncryptionTime: 0.000ns
         - TotalIntegrityCheckTime: 0.000ns
         - TotalReadBlockTime: 0.000ns
      EXCHANGE_NODE (id=4):(Total: 3s226ms, non-child: 0.000ns, % non-child: 0.00%)
        BytesReceived(500.000ms): 0, 0, 0, 0, 0, 0, 0
         - BytesReceived: 61.05 KB (62513)
         - ConvertRowBatchTime: 0.000ns
         - DeserializeRowBatchTimer: 0.000ns
         - FirstBatchArrivalWaitTime: 3s226ms
         - PeakMemoryUsage: 0
         - RowsReturned: 100 (100)
         - RowsReturnedRate: 30.00 /sec
         - SendersBlockedTimer: 0.000ns
         - SendersBlockedTotalTimer(*): 0.000ns
    Averaged Fragment F00:(Total: 3s001ms, non-child: 0.000ns, % non-child: 0.00%)
      split sizes:  min: 66.83 KB, max: 66.83 KB, avg: 66.83 KB, stddev: 0
      completion times: min:3s227ms  max:3s227ms  mean: 3s227ms  stddev:0.000ns
      execution rates: min:20.70 KB/sec  max:20.70 KB/sec  mean:20.70 KB/sec  stddev:0.00 /sec
      num instances: 1
       - AverageThreadTokens: 1.86
       - BloomFilterBytes: 1.00 MB (1048576)
       - PeakMemoryUsage: 5.07 MB (5320864)
       - PerHostPeakMemUsage: 108.65 MB (113924736)
       - PrepareTime: 38.999ms
       - RowsProduced: 1.02K (1024)
       - TotalCpuTime: 3s232ms
       - TotalNetworkReceiveTime: 2s940ms
       - TotalNetworkSendTime: 0.000ns
       - TotalStorageWaitTime: 13.999ms
      CodeGen:(Total: 262.997ms, non-child: 262.997ms, % non-child: 100.00%)
         - CodegenTime: 999.990us
         - CompileTime: 73.999ms
         - LoadTime: 0.000ns
         - ModuleBitcodeSize: 1.86 MB (1953028)
         - NumFunctions: 85 (85)
         - NumInstructions: 2.86K (2857)
         - OptimizationTime: 151.998ms
         - PrepareTime: 36.999ms
      DataStreamSender (dst_id=4):(Total: 999.990us, non-child: 999.990us, % non-child: 100.00%)
         - BytesSent: 61.05 KB (62513)
         - NetworkThroughput(*): 0.00 /sec
         - OverallThroughput: 59.62 MB/sec
         - RowsReturned: 1.02K (1024)
         - SerializeBatchTime: 999.990us
         - TransmitDataRPCTime: 0.000ns
         - UncompressedRowBatchSize: 185.83 KB (190290)
      HASH_JOIN_NODE (id=2):(Total: 3s001ms, non-child: 42.999ms, % non-child: 1.43%)
         - BuildPartitionTime: 1.999ms
         - BuildRows: 9.15K (9153)
         - BuildRowsPartitioned: 9.15K (9153)
         - BuildTime: 0.000ns
         - GetNewBlockTime: 0.000ns
         - HashBuckets: 16.38K (16384)
         - HashCollisions: 0 (0)
         - LargestPartitionPercent: 6 (6)
         - MaxPartitionLevel: 0 (0)
         - NumRepartitions: 0 (0)
         - PartitionsCreated: 16 (16)
         - PeakMemoryUsage: 3.29 MB (3445888)
         - PinTime: 0.000ns
         - ProbeRows: 1.02K (1024)
         - ProbeRowsPartitioned: 0 (0)
         - ProbeTime: 0.000ns
         - RowsReturned: 0 (0)
         - RowsReturnedRate: 0
         - SpilledPartitions: 0 (0)
         - UnpinTime: 0.000ns
        EXCHANGE_NODE (id=3):(Total: 2s941ms, non-child: 2s941ms, % non-child: 100.00%)
           - BytesReceived: 314.85 KB (322407)
           - ConvertRowBatchTime: 0.000ns
           - DeserializeRowBatchTimer: 0.000ns
           - FirstBatchArrivalWaitTime: 0.000ns
           - PeakMemoryUsage: 0
           - RowsReturned: 9.15K (9153)
           - RowsReturnedRate: 3.11 K/sec
           - SendersBlockedTimer: 0.000ns
           - SendersBlockedTotalTimer(*): 0.000ns
      HDFS_SCAN_NODE (id=1):(Total: 16.999ms, non-child: 16.999ms, % non-child: 100.00%)
         - AverageHdfsReadThreadConcurrency: 0.00
         - AverageScannerThreadConcurrency: 0.00
         - BytesRead: 128.51 KB (131593)
         - BytesReadDataNodeCache: 0
         - BytesReadLocal: 0
         - BytesReadRemoteUnexpected: 0
         - BytesReadShortCircuit: 0
         - DecompressionTime: 0.000ns
         - MaxCompressedTextFileLength: 0
         - NumColumns: 49 (49)
         - NumDisksAccessed: 0 (0)
         - NumRowGroups: 1 (1)
         - NumScannerThreadsStarted: 1 (1)
         - PeakMemoryUsage: 1.78 MB (1866400)
         - PerReadThreadRawHdfsThroughput: 4.18 MB/sec
         - RemoteScanRanges: 0 (0)
         - RowsRead: 1.50K (1500)
         - RowsReturned: 1.02K (1024)
         - RowsReturnedRate: 60.23 K/sec
         - ScanRangesComplete: 1 (1)
         - ScannerThreadsInvoluntaryContextSwitches: 0 (0)
         - ScannerThreadsTotalWallClockTime: 14.999ms
           - MaterializeTupleTime(*): 999.990us
           - ScannerThreadsSysTime: 0.000ns
           - ScannerThreadsUserTime: 2.216ms
         - ScannerThreadsVoluntaryContextSwitches: 27 (27)
         - TotalRawHdfsReadTime(*): 29.999ms
         - TotalReadThroughput: 0.00 /sec
        Filter 0 (1.00 MB):
           - Rows processed: 1.50K (1500)
           - Rows rejected: 3 (3)
           - Rows total: 1.50K (1500)
    Averaged Fragment F01:(Total: 3s191ms, non-child: 160.998ms, % non-child: 5.04%)
      split sizes:  min: 171.69 MB, max: 171.69 MB, avg: 171.69 MB, stddev: 0
      completion times: min:3s210ms  max:3s210ms  mean: 3s210ms  stddev:0.000ns
      execution rates: min:53.47 MB/sec  max:53.47 MB/sec  mean:53.47 MB/sec  stddev:0.00 /sec
      num instances: 1
       - AverageThreadTokens: 1.86
       - BloomFilterBytes: 0
       - PeakMemoryUsage: 106.05 MB (111206232)
       - PerHostPeakMemUsage: 108.65 MB (113924736)
       - PrepareTime: 33.999ms
       - RowsProduced: 9.15K (9153)
       - TotalCpuTime: 6s330ms
       - TotalNetworkReceiveTime: 0.000ns
       - TotalNetworkSendTime: 0.000ns
       - TotalStorageWaitTime: 36.999ms
      CodeGen:(Total: 51.999ms, non-child: 51.999ms, % non-child: 100.00%)
         - CodegenTime: 999.990us
         - CompileTime: 5.999ms
         - LoadTime: 0.000ns
         - ModuleBitcodeSize: 1.86 MB (1953028)
         - NumFunctions: 13 (13)
         - NumInstructions: 228 (228)
         - OptimizationTime: 11.999ms
         - PrepareTime: 33.999ms
      DataStreamSender (dst_id=3):(Total: 4.999ms, non-child: 4.999ms, % non-child: 100.00%)
         - BytesSent: 314.85 KB (322407)
         - NetworkThroughput(*): 153.74 MB/sec
         - OverallThroughput: 61.49 MB/sec
         - RowsReturned: 9.15K (9153)
         - SerializeBatchTime: 2.999ms
         - TransmitDataRPCTime: 1.999ms
         - UncompressedRowBatchSize: 769.68 KB (788150)
      HDFS_SCAN_NODE (id=0):(Total: 2s973ms, non-child: 2s973ms, % non-child: 100.00%)
         - AverageHdfsReadThreadConcurrency: 0.00
         - AverageScannerThreadConcurrency: 1.00
         - BytesRead: 171.79 MB (180132958)
         - BytesReadDataNodeCache: 0
         - BytesReadLocal: 0
         - BytesReadRemoteUnexpected: 0
         - BytesReadShortCircuit: 0
         - DecompressionTime: 130.998ms
         - MaxCompressedTextFileLength: 0
         - NumColumns: 8 (8)
         - NumDisksAccessed: 1 (1)
         - NumRowGroups: 1 (1)
         - NumScannerThreadsStarted: 1 (1)
         - PeakMemoryUsage: 106.05 MB (111196408)
         - PerReadThreadRawHdfsThroughput: 434.91 MB/sec
         - RemoteScanRanges: 0 (0)
         - RowsRead: 28.05M (28047320)
         - RowsReturned: 9.15K (9153)
         - RowsReturnedRate: 3.08 K/sec
         - ScanRangesComplete: 1 (1)
         - ScannerThreadsInvoluntaryContextSwitches: 112 (112)
         - ScannerThreadsTotalWallClockTime: 3s157ms
           - MaterializeTupleTime(*): 2s977ms
           - ScannerThreadsSysTime: 566.243ms
           - ScannerThreadsUserTime: 2s525ms
         - ScannerThreadsVoluntaryContextSwitches: 100 (100)
         - TotalRawHdfsReadTime(*): 394.996ms
         - TotalReadThroughput: 57.11 MB/sec
    Fragment F00:
      Instance dc4c084615fbf9bb:4261466f00000001 (host=mapr1:22000):(Total: 3s001ms, non-child: 0.000ns, % non-child: 0.00%)
        Hdfs split stats (<volume id>:<# splits>/<split lengths>): -1:1/66.83 KB
        Filter 0 arrival: 3s210ms
        MemoryUsage(500.000ms): 0, 3.05 MB, 3.14 MB, 3.13 MB, 3.09 MB, 3.12 MB, 3.19 MB
        ThreadUsage(500.000ms): 1, 2, 2, 2, 2, 2, 2
         - AverageThreadTokens: 1.86
         - BloomFilterBytes: 1.00 MB (1048576)
         - PeakMemoryUsage: 5.07 MB (5320864)
         - PerHostPeakMemUsage: 108.65 MB (113924736)
         - PrepareTime: 38.999ms
         - RowsProduced: 1.02K (1024)
         - TotalCpuTime: 3s232ms
         - TotalNetworkReceiveTime: 2s940ms
         - TotalNetworkSendTime: 0.000ns
         - TotalStorageWaitTime: 13.999ms
        CodeGen:(Total: 262.997ms, non-child: 262.997ms, % non-child: 100.00%)
           - CodegenTime: 999.990us
           - CompileTime: 73.999ms
           - LoadTime: 0.000ns
           - ModuleBitcodeSize: 1.86 MB (1953028)
           - NumFunctions: 85 (85)
           - NumInstructions: 2.86K (2857)
           - OptimizationTime: 151.998ms
           - PrepareTime: 36.999ms
        DataStreamSender (dst_id=4):(Total: 999.990us, non-child: 999.990us, % non-child: 100.00%)
           - BytesSent: 61.05 KB (62513)
           - NetworkThroughput(*): 0.00 /sec
           - OverallThroughput: 59.62 MB/sec
           - RowsReturned: 1.02K (1024)
           - SerializeBatchTime: 999.990us
           - TransmitDataRPCTime: 0.000ns
           - UncompressedRowBatchSize: 185.83 KB (190290)
        HASH_JOIN_NODE (id=2):(Total: 3s001ms, non-child: 42.999ms, % non-child: 1.43%)
          ExecOption: Build Side Codegen Enabled, Probe Side Codegen Enabled, Hash Table Construction Codegen Enabled, Join Build-Side Prepared Asynchronously, 1 of 1 Runtime Filter Published
           - BuildPartitionTime: 1.999ms
           - BuildRows: 9.15K (9153)
           - BuildRowsPartitioned: 9.15K (9153)
           - BuildTime: 0.000ns
           - GetNewBlockTime: 0.000ns
           - HashBuckets: 16.38K (16384)
           - HashCollisions: 0 (0)
           - LargestPartitionPercent: 6 (6)
           - MaxPartitionLevel: 0 (0)
           - NumRepartitions: 0 (0)
           - PartitionsCreated: 16 (16)
           - PeakMemoryUsage: 3.29 MB (3445888)
           - PinTime: 0.000ns
           - ProbeRows: 1.02K (1024)
           - ProbeRowsPartitioned: 0 (0)
           - ProbeTime: 0.000ns
           - RowsReturned: 0 (0)
           - RowsReturnedRate: 0
           - SpilledPartitions: 0 (0)
           - UnpinTime: 0.000ns
          EXCHANGE_NODE (id=3):(Total: 2s941ms, non-child: 999.990us, % non-child: 0.03%)
            BytesReceived(500.000ms): 70.08 KB, 127.46 KB, 162.66 KB, 230.08 KB, 301.42 KB, 312.19 KB
             - BytesReceived: 314.85 KB (322407)
             - ConvertRowBatchTime: 0.000ns
             - DeserializeRowBatchTimer: 0.000ns
             - FirstBatchArrivalWaitTime: 0.000ns
             - PeakMemoryUsage: 0
             - RowsReturned: 9.15K (9153)
             - RowsReturnedRate: 3.11 K/sec
             - SendersBlockedTimer: 0.000ns
             - SendersBlockedTotalTimer(*): 0.000ns
        HDFS_SCAN_NODE (id=1):(Total: 16.999ms, non-child: 16.999ms, % non-child: 100.00%)
          ExecOption: Expr Evaluation Codegen Disabled, PARQUET Codegen Enabled
          Hdfs split stats (<volume id>:<# splits>/<split lengths>): -1:1/66.83 KB
          Runtime filters: All filters arrived. Waited 0
          BytesRead(500.000ms): 0, 0, 0, 0, 0, 0
           - AverageHdfsReadThreadConcurrency: 0.00
           - AverageScannerThreadConcurrency: 0.00
           - BytesRead: 128.51 KB (131593)
           - BytesReadDataNodeCache: 0
           - BytesReadLocal: 0
           - BytesReadRemoteUnexpected: 0
           - BytesReadShortCircuit: 0
           - DecompressionTime: 0.000ns
           - MaxCompressedTextFileLength: 0
           - NumColumns: 49 (49)
           - NumDisksAccessed: 0 (0)
           - NumRowGroups: 1 (1)
           - NumScannerThreadsStarted: 1 (1)
           - PeakMemoryUsage: 1.78 MB (1866400)
           - PerReadThreadRawHdfsThroughput: 4.18 MB/sec
           - RemoteScanRanges: 0 (0)
           - RowsRead: 1.50K (1500)
           - RowsReturned: 1.02K (1024)
           - RowsReturnedRate: 60.23 K/sec
           - ScanRangesComplete: 1 (1)
           - ScannerThreadsInvoluntaryContextSwitches: 0 (0)
           - ScannerThreadsTotalWallClockTime: 14.999ms
             - MaterializeTupleTime(*): 999.990us
             - ScannerThreadsSysTime: 0.000ns
             - ScannerThreadsUserTime: 2.216ms
           - ScannerThreadsVoluntaryContextSwitches: 27 (27)
           - TotalRawHdfsReadTime(*): 29.999ms
           - TotalReadThroughput: 0.00 /sec
          Filter 0 (1.00 MB):
             - Rows processed: 1.50K (1500)
             - Rows rejected: 3 (3)
             - Rows total: 1.50K (1500)
    Fragment F01:
      Instance dc4c084615fbf9bb:4261466f00000002 (host=mapr1:22000):(Total: 3s191ms, non-child: 160.998ms, % non-child: 5.04%)
        Hdfs split stats (<volume id>:<# splits>/<split lengths>): -1:1/171.69 MB
        MemoryUsage(500.000ms): 0, 91.50 MB, 82.91 MB, 86.63 MB, 67.68 MB, 66.67 MB, 53.51 MB
        ThreadUsage(500.000ms): 1, 2, 2, 2, 2, 2, 2
         - AverageThreadTokens: 1.86
         - BloomFilterBytes: 0
         - PeakMemoryUsage: 106.05 MB (111206232)
         - PerHostPeakMemUsage: 108.65 MB (113924736)
         - PrepareTime: 33.999ms
         - RowsProduced: 9.15K (9153)
         - TotalCpuTime: 6s330ms
         - TotalNetworkReceiveTime: 0.000ns
         - TotalNetworkSendTime: 0.000ns
         - TotalStorageWaitTime: 36.999ms
        CodeGen:(Total: 51.999ms, non-child: 51.999ms, % non-child: 100.00%)
           - CodegenTime: 999.990us
           - CompileTime: 5.999ms
           - LoadTime: 0.000ns
           - ModuleBitcodeSize: 1.86 MB (1953028)
           - NumFunctions: 13 (13)
           - NumInstructions: 228 (228)
           - OptimizationTime: 11.999ms
           - PrepareTime: 33.999ms
        DataStreamSender (dst_id=3):(Total: 4.999ms, non-child: 4.999ms, % non-child: 100.00%)
           - BytesSent: 314.85 KB (322407)
           - NetworkThroughput(*): 153.74 MB/sec
           - OverallThroughput: 61.49 MB/sec
           - RowsReturned: 9.15K (9153)
           - SerializeBatchTime: 2.999ms
           - TransmitDataRPCTime: 1.999ms
           - UncompressedRowBatchSize: 769.68 KB (788150)
        HDFS_SCAN_NODE (id=0):(Total: 2s973ms, non-child: 2s973ms, % non-child: 100.00%)
          ExecOption: Expr Evaluation Codegen Enabled, PARQUET Codegen Enabled, Codegen enabled: 1 out of 1
          Hdfs split stats (<volume id>:<# splits>/<split lengths>): -1:1/171.69 MB
          Hdfs Read Thread Concurrency Bucket: 0:100% 1:0% 2:0% 3:0% 4:0% 5:0%
          File Formats: PARQUET/SNAPPY:8
          BytesRead(500.000ms): 88.45 MB, 112.45 MB, 136.45 MB, 152.66 MB, 168.66 MB, 171.79 MB
           - AverageHdfsReadThreadConcurrency: 0.00
           - AverageScannerThreadConcurrency: 1.00
           - BytesRead: 171.79 MB (180132958)
           - BytesReadDataNodeCache: 0
           - BytesReadLocal: 0
           - BytesReadRemoteUnexpected: 0
           - BytesReadShortCircuit: 0
           - DecompressionTime: 130.998ms
           - MaxCompressedTextFileLength: 0
           - NumColumns: 8 (8)
           - NumDisksAccessed: 1 (1)
           - NumRowGroups: 1 (1)
           - NumScannerThreadsStarted: 1 (1)
           - PeakMemoryUsage: 106.05 MB (111196408)
           - PerReadThreadRawHdfsThroughput: 434.91 MB/sec
           - RemoteScanRanges: 0 (0)
           - RowsRead: 28.05M (28047320)
           - RowsReturned: 9.15K (9153)
           - RowsReturnedRate: 3.08 K/sec
           - ScanRangesComplete: 1 (1)
           - ScannerThreadsInvoluntaryContextSwitches: 112 (112)
           - ScannerThreadsTotalWallClockTime: 3s157ms
             - MaterializeTupleTime(*): 2s977ms
             - ScannerThreadsSysTime: 566.243ms
             - ScannerThreadsUserTime: 2s525ms
           - ScannerThreadsVoluntaryContextSwitches: 100 (100)
           - TotalRawHdfsReadTime(*): 394.996ms
           - TotalReadThroughput: 57.11 MB/sec
1 ACCEPTED SOLUTION

Accepted Solutions
Highlighted

Re: Imapla query is slow, where is the bottleneck?

In the summary you can see that the average time in Node 0 is 2s973ms.The time in scan nodes is the time spent waiting for scanner threads to return a row batch. That looks like the bottleneck.

 

Drilling down into the profile for that node (HDFS_SCAN_NODE (id=0):(Total: 2s973ms, non-child: 2s973ms, % non-child: 100.00%)) we can see that it scanned a single Parquet file and spent almost all of that time materializing tuples (MaterializeTupleTime).

 

So there are two things going on:

  • Tuple materialisation is expensive.
  • There's no parallelism available because there's only a single parquet file in the partition. Usually for large tables having large files is a good thing, because there's less overhead, but in this case you're losing out on some parallelism.

We have a lot of changes in the pipeline at various stages to improve the first part - either by making it more efficient or by getting better at skipping over data. There's lots of room for improvement. Unfortunately that doesn't help you immediately.

View solution in original post

2 REPLIES 2
Highlighted

Re: Imapla query is slow, where is the bottleneck?

In the summary you can see that the average time in Node 0 is 2s973ms.The time in scan nodes is the time spent waiting for scanner threads to return a row batch. That looks like the bottleneck.

 

Drilling down into the profile for that node (HDFS_SCAN_NODE (id=0):(Total: 2s973ms, non-child: 2s973ms, % non-child: 100.00%)) we can see that it scanned a single Parquet file and spent almost all of that time materializing tuples (MaterializeTupleTime).

 

So there are two things going on:

  • Tuple materialisation is expensive.
  • There's no parallelism available because there's only a single parquet file in the partition. Usually for large tables having large files is a good thing, because there's less overhead, but in this case you're losing out on some parallelism.

We have a lot of changes in the pipeline at various stages to improve the first part - either by making it more efficient or by getting better at skipping over data. There's lots of room for improvement. Unfortunately that doesn't help you immediately.

View solution in original post

Highlighted

Re: Imapla query is slow, where is the bottleneck?

Explorer

Thank you sir for helping me walk through the profile. This is very informative. 

Don't have an account?
Coming from Hortonworks? Activate your account here