bigcyy opened a new issue, #3496:
URL: https://github.com/apache/hertzbeat/issues/3496
### Question
While recently reviewing the official Prometheus documentation, I noticed
the behavior of PromQL comparison operators (>, <, ==, >=, <=, !=) and found
two main inconsistencies when comparing it with the implementation in Hertzbeat.
1. Comparison Logic for Instant Vector vs. Scalar
According to the [official Prometheus
documentation](https://prometheus.io/docs/prometheus/latest/querying/operators/),
when an instant vector is compared to a scalar, the operation acts as a
filter. Specifically, PromQL iterates through each time series in the vector,
performs the comparison, and any time series for which the expression is true
is kept in the output. Series for which the result is false are removed.
However, I noticed that the behavior demonstrated in the calculate1 test
case in hertzbeat.warehouse.service.DataSourceServiceTest does not align with
this standard.
```java
void calculate1() {
List<Map<String, Object>> prometheusData = List.of(
new HashMap<>(Map.of("__value__", 100.0, "timestamp",
1343554, "instance", "node1")),
new HashMap<>(Map.of("__value__", 200.0, "timestamp",
1343555, "instance", "node2"))
);
QueryExecutor mockExecutor = Mockito.mock(QueryExecutor.class);
Mockito.when(mockExecutor.support("promql")).thenReturn(true);
Mockito.when(mockExecutor.execute(Mockito.anyString())).thenReturn(prometheusData);
dataSourceService.setExecutors(List.of(mockExecutor));
List<Map<String, Object>> result =
dataSourceService.calculate("promql", "node_cpu_seconds_total > 150");
assertEquals(2, result.size());
assertNull(result.get(0).get("__value__"));
assertEquals(200.0, result.get(1).get("__value__"));
}
```
Analysis:
Input: An instant vector containing two time series with values 100.0
and 200.0.
Query: node_cpu_seconds_total > 150.
Expected PromQL Behavior: Only the series with the value 200.0 satisfies
the condition. Therefore, the result should be a new vector containing only
that single series, and its size should be 1.
Behavior in Hertzbeat Test: The test asserts that result.size() is 2.
Instead of filtering out the non-matching series (__value__ of 100.0), it keeps
the series in the result set but sets its __value__ to null.
Core Question:
This "keep and nullify" behavior deviates from PromQL's standard
filtering model. Is this an intentional design choice in Hertzbeat? My main
question is whether the plan is to fully align with the Prometheus
specification in this regard. If not, it could cause confusion for users
familiar with standard PromQL.
2. Support for Comparison Operations on Range Vectors
The Prometheus documentation clearly states that logical and comparison
operators are only defined to work between instant vectors. They do not support
range vectors.
However, I noticed that DataSourceServiceTest contains test cases
involving comparisons between a range vector and a scalar, such as:
node_cpu_seconds_total{mode=\"user\"}[4m] <= 100
This is an invalid expression in standard PromQL. This seems to suggest
that Hertzbeat is extending PromQL syntax to support comparisons on range
vectors.
@tomsun28 @MasamiYui @Duansg
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail:
notifications-unsubscr...@hertzbeat.apache.org.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: notifications-unsubscr...@hertzbeat.apache.org
For additional commands, e-mail: notifications-h...@hertzbeat.apache.org
