fix : sorting rate metrics #102

sarthak77 · 2021-11-19T09:34:01Z

On the service (backend) screen page, there is a rate column - error/s. After we moved the avgrate functionality to query-service, the sorting on this column is broken.

The reason it was broken, that avgrate function takes two arguments (column name, size). The second size argument is missing in order by expression. Priori, it was working fine as avgrate was translated to sum at the gateway service layer. And sum function needs only single column name argument.

As these rate columns are per second rate, for now, defaulting to PT1S as size for avgrate function to fix the regression.

codecov · 2021-11-19T09:35:43Z

Codecov Report

Merging #102 (4d451f9) into main (cc5cdff) will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@             Coverage Diff              @@
##               main     #102      +/-   ##
============================================
+ Coverage     82.47%   82.48%   +0.01%     
- Complexity      419      421       +2     
============================================
  Files            37       37              
  Lines          1603     1604       +1     
  Branches        168      169       +1     
============================================
+ Hits           1322     1323       +1     
  Misses          207      207              
  Partials         74       74

Flag	Coverage Δ
integration	`82.48% <100.00%> (+0.01%)`	⬆️
unit	`80.29% <100.00%> (+0.07%)`	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...rvice/pinot/converters/PinotFunctionConverter.java	`94.87% <100.00%> (+0.06%)`	⬆️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cc5cdff...4d451f9. Read the comment docs.

kotharironak · 2021-11-19T09:42:04Z

...src/main/java/org/hypertrace/core/query/service/pinot/converters/PinotFunctionConverter.java

+    return function.getArgumentsList().stream().filter(e -> e.hasColumnIdentifier()).findFirst();
+  }
+
+  private String getAvgRateLiteral(Function function) {


getAvgRateLiteral -> getLiteralForAvgRate?

getRateIntervalForAvgRate ?

kotharironak · 2021-11-19T09:42:49Z

query-service-impl/src/test/resources/service_request_handler.conf

@@ -0,0 +1,28 @@
+{


Can you add tests ?

kotharironak

lgtm

avinashkolluru · 2021-11-19T12:55:26Z

...src/main/java/org/hypertrace/core/query/service/pinot/converters/PinotFunctionConverter.java

@@ -96,6 +95,18 @@ private String functionToStringForAvgRate(
        columnName, (double) aggregateIntervalInSeconds / rateIntervalInSeconds);
  }

+  private Optional<Expression> getColumnNameForAvgRate(Function function) {


Why are we returning an Optional here if the caller is forcefully doing get()?
Optional here makes sense if the caller needs to deal with the value being present or not.

Agreed. Also, this isn't returning a column name, it's returning an expression. If I sent my args out of order, this would still work which is not what we want - we just want backwards compatibility. I would just change the old line 84 to default if empty, and leave the rest as is.

aaron-steinfeld · 2021-11-19T13:23:30Z

...src/main/java/org/hypertrace/core/query/service/pinot/converters/PinotFunctionConverter.java

@@ -80,9 +80,8 @@ private String functionToStringForAvgRate(
      java.util.function.Function<Expression, String> argumentConverter,
      ExecutionContext executionContext) {

-    String columnName = argumentConverter.apply(function.getArgumentsList().get(0));
-    String rateIntervalInIso =
-        function.getArgumentsList().get(1).getLiteral().getValue().getString();


That is, change this to something equivalent to

function.getArgumentsList().size() == 2 ? function.getArgumentsList().get(1).getLiteral().getValue().getString() : DEFAULT_AVG_RATE_SIZE;

made the change

avinashkolluru · 2021-11-19T15:00:57Z

...src/main/java/org/hypertrace/core/query/service/pinot/converters/PinotFunctionConverter.java

+    return function.getArgumentsList().stream().filter(e -> e.hasColumnIdentifier()).findFirst();
+  }
+
+  private String getRateIntervalForAvgRate(Function function) {


We have integration-tests in query-service. We should definitely add one for this case.

We already have two integration tests for testing the average rate changes. Should I add for this specific case also where average rate is in order-by expression ?

Yes. We should definitely add one.

aaron-steinfeld · 2021-11-19T16:09:26Z

...c/test/java/org/hypertrace/core/query/service/pinot/QueryRequestToPinotSQLConverterTest.java

@@ -1120,14 +1242,19 @@ private QueryRequest buildSimpleQueryWithFilter(Filter filter) {
    return builder.build();
  }

-  private void assertPQLQuery(QueryRequest queryRequest, String expectedQuery) {
+  private void assertPQLQuery(
+      QueryRequest queryRequest, String expectedQuery, ViewDefinition viewDefinition) {


if we're going to require passing this in (which is fine), we shouldn't make it a global static.

This was because we need service view definition for testing avgrate changes. Rest all use this same view definition.

aaron-steinfeld · 2021-11-19T16:09:39Z

...c/test/java/org/hypertrace/core/query/service/pinot/QueryRequestToPinotSQLConverterTest.java

+  private void assertPQLQuery(
+      QueryRequest queryRequest, String expectedQuery, ViewDefinition viewDefinition) {
+
+    when(mockingExecutionContext.getTenantId()).thenReturn("__default");


An assertion shouldn't do mocking of its own. I would guess these are not used by most test cases, which means mockito would be erroring out unless we're running in lenient mode (I'm guessing we are). We don't need to switch lenient off as part of this change, but we should be strict for any new tests we write.

https://javadoc.io/static/org.mockito/mockito-core/3.2.4/org/mockito/quality/Strictness.html#STRICT_STUBS

getTenantId() is used by most of the tests.

moved others to the respective test.

we shouldn't share the state though - that's the same problem you ran into with view definition. Have each test pass in their own execution context, same as view def, then they can mock that locally, according to their test parameters. For backwards compat, you could overload the assert method with the existing behavior if you want - that way no existing tests hve to be modified. In the current for, we have a mix of both things we don't want - shared state (including newly introduced shared state) and modifying the existing tests.

sarthak77 · 2021-11-19T16:16:31Z

...c/test/java/org/hypertrace/core/query/service/pinot/QueryRequestToPinotSQLConverterTest.java

+  }
+
+  @Test
+  public void testQueryWithAverageRateInOrderBy() {


@aaron-steinfeld @avinashkolluru the integration test would be exactly the same as this. Should I keep this test also or remove this one?

I think we should keep this as well. It is checking the expected PQL query. The integration test will be checking the output of the API.

aaron-steinfeld

approving to unblock the bug fix, but we should get the newly added tests cleaned up here or in a followup.

aaron-steinfeld · 2021-11-21T23:06:25Z

...c/test/java/org/hypertrace/core/query/service/pinot/QueryRequestToPinotSQLConverterTest.java

    Mockito.when(connection.prepareStatement(any(Request.class))).thenCallRealMethod();
+    mockingExecutionContext = mock(ExecutionContext.class);
+    when(mockingExecutionContext.getTenantId()).thenReturn("__default");
+    when(this.mockingExecutionContext.getTimeSeriesPeriod()).thenReturn(Optional.empty());


this mock is not relevant to every test and would fail in strict mode. please move it such that we only mock things that are relevant to the test.

For future tests (beyond the scope of this PR), make sure to always run in strict mode. The easiest way to do that is probably use the junit extension -

@ExtendWith(MockitoExtension.class)

github-actions · 2021-11-22T03:23:46Z

Unit Test Results

  22 files ±0   22 suites ±0 8s ⏱️ -1s
126 tests +2 126 ✔️ +2 0 💤 ±0 0 ❌ ±0

Results for commit 7df2959. ± Comparison against base commit cc5cdff.

sarthak77 · 2021-11-22T08:58:52Z

Raised a follow-up PR for fixing the unit tests : #103

kotharironak and others added 2 commits November 19, 2021 13:42

fix : defaulting avg_rate argument to PT1S

23d3d51

Merge branch 'main' into fix-order-by-expression

c9a3ee8