Performance improvement on codec mapping

[frdeboffles]

Feature Request

While using this driver for fetching many rows from our table I noticed it was taking quite some time to build the Row objects. I hooked a profiler to the code and noticed it was spending quite some time for each row on the codec.canDecode method.
I wrote this little perf test to validate this finding:

  @Test
  public void perfTestDecode() {
    DefaultCodecs codecs = new DefaultCodecs(TEST);
    long t = System.currentTimeMillis();
    for (int i = 0; i < 100000; i++) {
      for (int c = 0; c < 20; c++) {
        assertThat(
            codecs.decode(
                TEST.buffer(4).writeInt(100), INT4.getObjectId(), FORMAT_BINARY, Integer.class))
            .isEqualTo(100);
        assertThat(
            codecs.decode(
                ByteBufUtils.encode(TEST, "100"), INT2.getObjectId(), FORMAT_TEXT, Short.class))
            .isEqualTo((short) 100);
        assertThat(
            codecs.decode(
                ByteBufUtils.encode(TEST, "test"),
                VARCHAR.getObjectId(),
                FORMAT_TEXT,
                String.class))
            .isEqualTo("test");
        assertThat(
            codecs.decode(
                ByteBufUtils.encode(TEST, "2018-11-04 15:35:00.847108"),
                TIMESTAMP.getObjectId(),
                FORMAT_TEXT,
                LocalDateTime.class))
            .isInstanceOf(LocalDateTime.class);
        assertThat(codecs.decode(ByteBufUtils.encode(TEST, "{100,200}"), INT2_ARRAY.getObjectId(), FORMAT_TEXT, Object.class)).isEqualTo(new Short[]{100, 200});
        assertThat(codecs.decode(ByteBufUtils.encode(TEST, "{100,200}"), INT4_ARRAY.getObjectId(), FORMAT_TEXT, Object.class)).isEqualTo(new Integer[]{100, 200});
      }
    }
    System.out.println("Run in " + (System.currentTimeMillis() - t) + "ms");
  }

Running this on my machine results in:

Run in 13231ms

NOTE That running the same example from 0.8.7.RELEASE results in

Run in 10389ms

So it seems like the decoding performance went down between this release and the current master (at commit 9c773c8)

Is your feature request related to a problem? Please describe

I would rather use this driver and webflux than going the web mvc route and jdbc.

Describe the solution you'd like

I think the class DefaultCodecs would benefits from some simple caching mechanism.
I did implement a simple caching (PR to come) and the results of the above test comes to:

Run in 7294ms

Describe alternatives you've considered

I haven't found alternatives

Teachability, Documentation, Adoption, Migration Strategy

NA

[mp911de]

Thanks a lot for your suggestion. I had something like this on my mind since iterations aren't really efficient to determine a codec.

[protyay]

@mp911de Interested to work in this project. Are you accepting new contributors ?

[mp911de]

Sure. Feel free to pick an issue, submit a pull request or participate in issue discussions.