|
| 1 | +# Every rule in the JMX configuration is covered by one expected metric. |
| 2 | + |
| 3 | +import re |
| 4 | +import sys |
| 5 | +import logging |
| 6 | + |
| 7 | +import requests |
| 8 | + |
| 9 | + |
| 10 | +def check_metrics( |
| 11 | + namespace: str, role: str, port: int, expected_metrics: list[str] |
| 12 | +) -> None: |
| 13 | + response: requests.Response = requests.get( |
| 14 | + f"http://hdfs-{role}-default-0.hdfs-{role}-default.{namespace}.svc.cluster.local:{port}/metrics", |
| 15 | + timeout=10, |
| 16 | + ) |
| 17 | + assert response.ok, "Requesting metrics failed" |
| 18 | + |
| 19 | + for metric in expected_metrics: |
| 20 | + assert ( |
| 21 | + re.search(f"^{metric}", response.text, re.MULTILINE) is not None |
| 22 | + ), f"Metric '{metric}' not found for {role}" |
| 23 | + |
| 24 | + |
| 25 | +def check_namenode_metrics( |
| 26 | + namespace: str, |
| 27 | + product_version: str, |
| 28 | +) -> None: |
| 29 | + expected_metrics: list[str] = [ |
| 30 | + # Kind "MetricsSystem" |
| 31 | + 'hadoop_namenode_num_active_sources{kind="MetricsSystem",role="NameNode",service="HDFS",sub="Stats"}', |
| 32 | + # Attribute "Total" |
| 33 | + 'hadoop_namenode_total{kind="NameNodeInfo",role="NameNode",service="HDFS"}', |
| 34 | + # Counter suffixed with "_total" |
| 35 | + 'hadoop_namenode_files_total{kind="FSNamesystem",role="NameNode",service="HDFS"}', |
| 36 | + # Metric suffixed with "_created" |
| 37 | + 'hadoop_namenode_files_created_{kind="NameNodeActivity",role="NameNode",service="HDFS"}', |
| 38 | + # Non-special metric |
| 39 | + 'hadoop_namenode_files_deleted{kind="NameNodeActivity",role="NameNode",service="HDFS"}', |
| 40 | + ] |
| 41 | + |
| 42 | + if product_version in ["3.3.4", "3.3.6"]: |
| 43 | + # Log counters were removed in 3.4.0 (HADOOP-17524). |
| 44 | + expected_metrics.extend( |
| 45 | + [ |
| 46 | + # Metric suffixed with "_info" |
| 47 | + 'hadoop_namenode_log_info_{kind="JvmMetrics",role="NameNode",service="HDFS"}', |
| 48 | + ] |
| 49 | + ) |
| 50 | + |
| 51 | + check_metrics(namespace, "namenode", 8183, expected_metrics) |
| 52 | + |
| 53 | + |
| 54 | +def check_datanode_metrics( |
| 55 | + namespace: str, |
| 56 | + product_version: str, |
| 57 | +) -> None: |
| 58 | + expected_metrics: list[str] = [ |
| 59 | + # Kind "MetricsSystem" |
| 60 | + 'hadoop_datanode_num_active_sources{kind="MetricsSystem",role="DataNode",service="HDFS",sub="Stats"}', |
| 61 | + # Kind "FSDatasetState" suffixed with "_total" |
| 62 | + 'hadoop_datanode_estimated_capacity_lost_total{fsdatasetid=".+",kind="FSDatasetState",role="DataNode",service="HDFS"}', |
| 63 | + # Kind "FSDatasetState" |
| 64 | + 'hadoop_datanode_capacity{fsdatasetid=".+",kind="FSDatasetState",role="DataNode",service="HDFS"}', |
| 65 | + # Kind "DataNodeActivity" suffixed with "_info" |
| 66 | + 'hadoop_datanode_blocks_get_local_path_info_{host="hdfs-datanode-default-0\\.hdfs-datanode-default\\..+\\.svc\\.cluster\\.local",kind="DataNodeActivity",port="9866",role="DataNode",service="HDFS"}', |
| 67 | + # Kind "DataNodeActivity" |
| 68 | + 'hadoop_datanode_blocks_read{host="hdfs-datanode-default-0\\.hdfs-datanode-default\\..+\\.svc\\.cluster\\.local",kind="DataNodeActivity",port="9866",role="DataNode",service="HDFS"}', |
| 69 | + # Counter suffixed with "_total" |
| 70 | + 'hadoop_datanode_estimated_capacity_lost_total{kind="FSDatasetState",role="DataNode",service="HDFS"}', |
| 71 | + # Non-special metric |
| 72 | + 'hadoop_datanode_gc_count{kind="JvmMetrics",role="DataNode",service="HDFS"}', |
| 73 | + ] |
| 74 | + |
| 75 | + if product_version in ["3.3.4", "3.3.6"]: |
| 76 | + # Log counters were removed in 3.4.0 (HADOOP-17524). |
| 77 | + expected_metrics.extend( |
| 78 | + [ |
| 79 | + # Metric suffixed with "_info" |
| 80 | + 'hadoop_datanode_log_info_{kind="JvmMetrics",role="DataNode",service="HDFS"}', |
| 81 | + ] |
| 82 | + ) |
| 83 | + |
| 84 | + check_metrics(namespace, "datanode", 8082, expected_metrics) |
| 85 | + |
| 86 | + |
| 87 | +def check_journalnode_metrics( |
| 88 | + namespace: str, |
| 89 | + product_version: str, |
| 90 | +) -> None: |
| 91 | + expected_metrics: list[str] = [ |
| 92 | + # Kind "MetricsSystem" |
| 93 | + 'hadoop_journalnode_num_active_sources{kind="MetricsSystem",role="JournalNode",service="HDFS",sub="Stats"}', |
| 94 | + # Non-special metric |
| 95 | + 'hadoop_journalnode_bytes_written{kind="Journal-hdfs",role="JournalNode",service="HDFS"}', |
| 96 | + ] |
| 97 | + |
| 98 | + if product_version in ["3.3.4", "3.3.6"]: |
| 99 | + # Log counters were removed in 3.4.0 (HADOOP-17524). |
| 100 | + expected_metrics.extend( |
| 101 | + [ |
| 102 | + # Metric suffixed with "_info" |
| 103 | + 'hadoop_journalnode_log_info_{kind="JvmMetrics",role="JournalNode",service="HDFS"}', |
| 104 | + ] |
| 105 | + ) |
| 106 | + |
| 107 | + check_metrics(namespace, "journalnode", 8081, expected_metrics) |
| 108 | + |
| 109 | + |
| 110 | +if __name__ == "__main__": |
| 111 | + namespace_arg: str = sys.argv[1] |
| 112 | + product_version_arg: str = sys.argv[2] |
| 113 | + |
| 114 | + logging.basicConfig( |
| 115 | + level="DEBUG", |
| 116 | + format="%(asctime)s %(levelname)s: %(message)s", |
| 117 | + stream=sys.stdout, |
| 118 | + ) |
| 119 | + |
| 120 | + check_namenode_metrics(namespace_arg, product_version_arg) |
| 121 | + check_datanode_metrics(namespace_arg, product_version_arg) |
| 122 | + check_journalnode_metrics(namespace_arg, product_version_arg) |
| 123 | + |
| 124 | + print("All expected metrics found") |
0 commit comments