CODEWITHSUNDEEP
javapostgresqljdbctimescaledb
pgPulp
pgPulp

Reputation: 31

TimescaleDB slower than plain postgresql 10 for insert

I am inserting 1m rows into a test table with timescale using JDBC and the performance seems to be about half that of plain postgresql. Timescale tuning was done by taking all values suggested by the timescale-tune utility. What am I doing wrong?

   private static void writeTable(String sql, int count, int commitCount,
   Connection conn) throws Exception
   {
       conn.setAutoCommit(false);
       long start = System.currentTimeMillis();
       long t = start;
       PreparedStatement stmt = conn.prepareStatement(sql);
       for(int i = 0; i < count; i++)
       {
           stmt.setTimestamp(1, new Timestamp(t));
           stmt.setDouble(2, 10.9);
           stmt.addBatch();
           t ++;
           if(commitCount != -1 && ((i + 1) % commitCount) == 0)
           {
               stmt.executeBatch();
               conn.commit();
           }
       }
       stmt.executeBatch();
       stmt.close();
       conn.commit();
       conn.close();
       long diff = System.currentTimeMillis() - start;
       System.out.println("Count      : " + count);
       System.out.println("Total Time : " + diff);
       System.out.println("Writes/Sec : " + ((count * 1000) / diff));
   }

Query: INSERT INTO kt_device_info (di_device_id, di_time, di_value) VALUES (1,?,?)

Table:
CREATE TABLE kt_device (
    id              BIGINT PRIMARY KEY,
    d_name          TEXT
);

insert into kt_device(id, d_name) values (1, 'dev-1');

CREATE TABLE kt_device_info (
    di_device_id    BIGINT REFERENCES  kt_device NOT NULL,
    di_time         TIMESTAMPTZ NOT NULL,
    di_value        DOUBLE PRECISION  NULL
);

SELECT create_hypertable('kt_device_info', 'di_time');

Timescale : Count : 1000000 Total Time : 42026 Writes/Sec : 23794

Postgres 10: Count : 1000000 Total Time : 22573 Writes/Sec : 44300

PostgreSQL 10.10 (Ubuntu 10.10-1.pgdg16.04+1) on x86_64-pc-linux-gnu, compiled by gcc (Ubuntu 5.4.0-6ubuntu1~16.04.11) 5.4.0 20160609, 64-bit

timescaledb | 1.4.2 | public | Enables scalable inserts and complex queries for time-series data

Hardware: Intel(R) Core(TM) i7-4702MQ CPU @ 2.20GHz, 16GB Memory

Chunks:
SELECT show_chunks('kt_device_info');
              show_chunks               
----------------------------------------
 _timescaledb_internal._hyper_7_7_chunk
(1 row)

Upvotes: 3

Views: 1746

Answers (1)

saule_goode
saule_goode

Reputation: 11

Looking at your code, you are creating timestamps that are milliseconds apart. That would explain why you only have one chunk. The default chunk size is 7 days. In this case you probably want to set the partitions smaller to something like a few seconds. You can change the chunk size with: SELECT set_chunk_time_interval('kt_device_info', 4000);

Upvotes: 1

Related Questions