[BACKPORT][Metric] Fix prometheus metric backend (#3124) (#3127)

Co-authored-by: bermaker <[email protected]>

docs/source/development/metrics.rst

@@ -59,8 +59,95 @@ Mars metrics support three different backends:
 * ``prometheus`` is an open-source systems monitoring and alerting toolkit.
 * ``ray`` is a metric backend which just runs on ray engine.
 
-We can choose a metric backend by configuring ``metrics.backend`` in
-``mars/deploy/oscar/base_config.yml`` or its descendant files.
+Console
+````````````````
+
+The default metric backend is ``console``. It just logs the value when log level
+is ``debug``.
+
+Prometheus
+````````````````
+
+Firstly, we should download Prometheus. For details, please refer to
+`Prometheus Getting Started
+<https://prometheus.io/docs/prometheus/latest/getting_started/>`_.
+
+Secondly, we can new a Mars session by configuring Prometheus backend as follows:
+
+.. code-block:: python
+
+    In [1]: import mars
+
+    In [2]: session = mars.new_session(
+       ...:     n_worker=1,
+       ...:     n_cpu=2,
+       ...:     web=True,
+       ...:     config={"metrics.backend": "prometheus"}
+       ...: )
+    Finished startup prometheus http server and port is 15768
+    Finished startup prometheus http server and port is 44303
+    Finished startup prometheus http server and port is 63391
+    Finished startup prometheus http server and port is 13722
+    Web service started at http://0.0.0.0:15518
+
+Thirdly, we should config Prometheus, more configurations please refer to
+`Prometheus Configuration
+<https://prometheus.io/docs/prometheus/latest/configuration/configuration/>`_.
+
+.. code-block:: yaml
+
+    scrape_configs:
+      - job_name: 'mars'
+
+        scrape_interval: 5s
+
+        static_configs:
+          - targets: ['localhost:15768', 'localhost:44303', 'localhost:63391', 'localhost:13722']
+
+
+Then start Prometheus:
+
+.. code-block:: shell
+
+    $ prometheus --config.file=promconfig.yaml
+    level=info ts=2022-06-07T13:05:01.484Z caller=main.go:296 msg="no time or size retention was set so using the default time retention" duration=15d
+    level=info ts=2022-06-07T13:05:01.484Z caller=main.go:332 msg="Starting Prometheus" version="(version=2.13.1, branch=non-git, revision=non-git)"
+    level=info ts=2022-06-07T13:05:01.484Z caller=main.go:333 build_context="(go=go1.13.1, [email protected], date=20191018-01:13:04)"
+    level=info ts=2022-06-07T13:05:01.485Z caller=main.go:334 host_details=(darwin)
+    level=info ts=2022-06-07T13:05:01.485Z caller=main.go:335 fd_limits="(soft=256, hard=unlimited)"
+    level=info ts=2022-06-07T13:05:01.485Z caller=main.go:336 vm_limits="(soft=unlimited, hard=unlimited)"
+    level=info ts=2022-06-07T13:05:01.487Z caller=main.go:657 msg="Starting TSDB ..."
+    level=info ts=2022-06-07T13:05:01.488Z caller=web.go:450 component=web msg="Start listening for connections" address=0.0.0.0:9090
+    level=info ts=2022-06-07T13:05:01.494Z caller=head.go:514 component=tsdb msg="replaying WAL, this may take awhile"
+    level=info ts=2022-06-07T13:05:01.495Z caller=head.go:562 component=tsdb msg="WAL segment loaded" segment=0 maxSegment=1
+    level=info ts=2022-06-07T13:05:01.495Z caller=head.go:562 component=tsdb msg="WAL segment loaded" segment=1 maxSegment=1
+    level=info ts=2022-06-07T13:05:01.497Z caller=main.go:672 fs_type=1a
+    level=info ts=2022-06-07T13:05:01.497Z caller=main.go:673 msg="TSDB started"
+    level=info ts=2022-06-07T13:05:01.497Z caller=main.go:743 msg="Loading configuration file" filename=promconfig_mars.yaml
+    level=info ts=2022-06-07T13:05:01.501Z caller=main.go:771 msg="Completed loading of configuration file" filename=promconfig_mars.yaml
+    level=info ts=2022-06-07T13:05:01.501Z caller=main.go:626 msg="Server is ready to receive web requests."
+
+Fourthly, run a Mars task:
+
+.. code-block:: python
+
+    In [3]: import numpy as np
+
+    In [4]: import mars.dataframe as md
+
+    In [5]: df1 = md.DataFrame(np.random.randint(0, 3, size=(10, 4)),
+       ...:                    columns=list('ABCD'), chunk_size=5)
+       ...: df2 = md.DataFrame(np.random.randint(0, 3, size=(10, 4)),
+       ...:                    columns=list('ABCD'), chunk_size=5)
+       ...:
+       ...: r = md.merge(df1, df2, on='A').execute()
+
+Finally, we can check metrics in Prometheus web http://localhost:9090.
+
+Ray
+````````````````
+
+We could config ``metrics.backend`` when creating a Ray cluster or new a session.
 
 Metrics Naming Convention
 ------------------

docs/source/locale/zh_CN/LC_MESSAGES/development/metrics.po

@@ -8,7 +8,7 @@ msgid ""
 msgstr ""
 "Project-Id-Version: mars 0.9.0rc2+18.g21929ced5\n"
 "Report-Msgid-Bugs-To: \n"
-"POT-Creation-Date: 2022-04-24 12:19+0800\n"
+"POT-Creation-Date: 2022-06-08 14:41+0800\n"
 "PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\n"
 "Last-Translator: FULL NAME <EMAIL@ADDRESS>\n"
 "Language-Team: LANGUAGE <[email protected]>\n"
@@ -53,8 +53,8 @@ msgstr "``Meter`` 是一组事件发生的速率。 我们可以将其用作 qps
 
 #: ../../source/development/metrics.rst:16
 msgid ""
-"``Histogram`` is a type of statistics which records the average value of"
-" a window data."
+"``Histogram`` is a type of statistics which records the average value of "
+"a window data."
 msgstr "``Histogram`` 是一种统计类型，它记录窗口数据的平均值。"
 
 #: ../../source/development/metrics.rst:18
@@ -66,8 +66,9 @@ msgid ""
 "**Note**: If ``tag_keys`` is declared, ``tags`` must be specified when "
 "invoking ``record`` method and tags' keys must be consistent with "
 "``tag_keys``."
-msgstr "**注意**：如果声明了 ``tag_keys``，调用 ``record`` 方法时必须指定 ``tags`` "
-"参数，并且 ``tags`` 的 keys 必须跟 ``tag_keys`` 保持一致。"
+msgstr ""
+"**注意**：如果声明了 ``tag_keys``，调用 ``record`` 方法时必须指定 ``tags`` 参数，并且 ``tags`` 的"
+" keys 必须跟 ``tag_keys`` 保持一致。"
 
 #: ../../source/development/metrics.rst:54
 msgid "Three different Backends"
@@ -89,40 +90,93 @@ msgstr "``prometheus`` 一个开源系统监控和报警工具包。"
 msgid "``ray`` is a metric backend which just runs on ray engine."
 msgstr "``ray`` 是一种运行在 ray 引擎上的 metric 后端。"
 
-#: ../../source/development/metrics.rst:62
+#: ../../source/development/metrics.rst:63
+msgid "Console"
+msgstr ""
+
+#: ../../source/development/metrics.rst:65
+msgid ""
+"The default metric backend is ``console``. It just logs the value when "
+"log level is ``debug``."
+msgstr "默认的 metric 后端是 ``console``. 它只是在日志级别为 ``debug`` 时打印出 metric 的值。"
+
+#: ../../source/development/metrics.rst:69
+msgid "Prometheus"
+msgstr ""
+
+#: ../../source/development/metrics.rst:71
+msgid ""
+"Firstly, we should download Prometheus. For details, please refer to "
+"`Prometheus Getting Started "
+"<https://prometheus.io/docs/prometheus/latest/getting_started/>`_."
+msgstr ""
+"首先，我们需要下载 Prometheus。具体的可以参考 `Prometheus Getting Started "
+"<https://prometheus.io/docs/prometheus/latest/getting_started/>`_."
+
+#: ../../source/development/metrics.rst:75
 msgid ""
-"We can choose a metric backend by configuring ``metrics.backend`` in "
-"``mars/deploy/oscar/base_config.yml`` or its descendant files."
-msgstr "我们可以通过配置 ``mars/deploy/oscar/base_config.yml`` 或它的继承文件中的 "
-"``metrics.backend`` 来选择一种 metric 后端。"
+"Secondly, we can new a Mars session by configuring Prometheus backend as "
+"follows:"
+msgstr "其次，我们可以如下配置 Prometheus 后端来启动一个 Mars session："
 
-#: ../../source/development/metrics.rst:66
+#: ../../source/development/metrics.rst:93
+msgid ""
+"Thirdly, we should config Prometheus, more configurations please refer to"
+" `Prometheus Configuration "
+"<https://prometheus.io/docs/prometheus/latest/configuration/configuration/>`_."
+msgstr ""
+"第三，我们要配置 Prometheus，更多的配置可以参考  `Prometheus Configuration "
+"<https://prometheus.io/docs/prometheus/latest/configuration/configuration/>`_."
+
+#: ../../source/development/metrics.rst:108
+msgid "Then start Prometheus:"
+msgstr "接着，启动 Prometheus："
+
+#: ../../source/development/metrics.rst:130
+msgid "Fourthly, run a Mars task:"
+msgstr "第四，执行一个 Mars task："
+
+#: ../../source/development/metrics.rst:145
+msgid "Finally, we can check metrics in Prometheus web http://localhost:9090."
+msgstr "最后，我们可以在 Prometheus 的网页端 http://localhost:9090 查看 metrics。"
+
+#: ../../source/development/metrics.rst:148
+msgid "Ray"
+msgstr ""
+
+#: ../../source/development/metrics.rst:150
+msgid ""
+"We could config ``metrics.backend`` when creating a Ray cluster or new a "
+"session."
+msgstr "我们可以在创建 Ray cluster 时或新建 session 时配置 ``metrics.backend``。"
+
+#: ../../source/development/metrics.rst:153
 msgid "Metrics Naming Convention"
 msgstr "Metrics 命名约定"
 
-#: ../../source/development/metrics.rst:68
+#: ../../source/development/metrics.rst:155
 msgid "We propose a naming convention for metrics as follows:"
 msgstr "我们提出一种如下的 metrics 命名约定："
 
-#: ../../source/development/metrics.rst:70
+#: ../../source/development/metrics.rst:157
 msgid "``namespace.[component].metric_name[_units]``"
 msgstr ""
 
-#: ../../source/development/metrics.rst:72
+#: ../../source/development/metrics.rst:159
 msgid "``namespace`` could be ``mars``."
 msgstr "``namespace`` 可以是 ``mars``。"
 
-#: ../../source/development/metrics.rst:73
-msgid "``component`` could be `supervisor`, `worker` or `band` etc, and can be "
+#: ../../source/development/metrics.rst:160
+msgid ""
+"``component`` could be `supervisor`, `worker` or `band` etc, and can be "
 "omitted."
 msgstr "``component`` 可以是 `supervisor`，`worker` 或 `band` 等等，也可以省略这个参数。"
 
-#: ../../source/development/metrics.rst:74
+#: ../../source/development/metrics.rst:161
 msgid ""
 "``units`` is the metric unit which may be seconds when recording time, or"
 " ``_count`` when metric type is ``Counter``, ``_number`` when metric type"
 " is ``Gauge`` if there is no suitable unit."
-msgstr "``units`` 是 metric 的单位，当记录的是时间时，可以用 seconds，当没有合适的单位"
-"时，``Counter`` 类型的 metric 可以用 ``_count``，``Gauge`` 类型的 metric 可以用 "
-"``_number``。"
-
+msgstr ""
+"``units`` 是 metric 的单位，当记录的是时间时，可以用 seconds，当没有合适的单位时，``Counter`` 类型的 "
+"metric 可以用 ``_count``，``Gauge`` 类型的 metric 可以用 ``_number``。"

mars/metrics/__init__.py

@@ -13,5 +13,5 @@
 # limitations under the License.
 
 from .api import Metrics
-from .api import init_metrics
+from .api import init_metrics, shutdown_metrics
 from .api import record_time_cost_percentile, Percentile

mars/metrics/api.py

@@ -26,6 +26,7 @@
 
 logger = logging.getLogger(__name__)
 
+_init = False
 _metric_backend = "console"
 _backends_cls = {
     "console": console_metric,
@@ -35,6 +36,10 @@
 
 
 def init_metrics(backend="console", config: Dict[str, Any] = None):
+    global _init
+    if _init is True:
+        return
+
     backend = backend or "console"
     if backend not in _backends_cls:
         raise NotImplementedError(f"Do not support metric backend {backend}")
@@ -43,17 +48,29 @@ def init_metrics(backend="console", config: Dict[str, Any] = None):
     if _metric_backend == "prometheus":
         try:
             from prometheus_client import start_http_server
+            from ..utils import get_next_port
 
             port = config.get("port", 0) if config else 0
+            port = port or get_next_port()
             start_http_server(port)
-            logger.info("Finished startup prometheus http server and port is %d", port)
+            logger.warning(
+                "Finished startup prometheus http server and port is %d", port
+            )
         except ImportError:
             logger.warning(
                 "Failed to start prometheus http server because there is no prometheus_client"
             )
+    _init = True
     logger.info("Finished initialize the metrics with backend %s", _metric_backend)
 
 
+def shutdown_metrics():
+    global _metric_backend
+    _metric_backend = "console"
+    global _init
+    _init = False
+
+
 class Metrics:
     """
     A factory to generate different types of metrics.

mars/metrics/backends/prometheus/prometheus_metric.py

@@ -12,6 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import os
+import socket
+
 from typing import Optional, Dict
 
 from ....utils import lazy_import
@@ -28,16 +31,28 @@
 
 class PrometheusMetricMixin(AbstractMetric):
     def _init(self):
-        self._metric = (
-            pc.Gauge(self._name, self._description, self._tag_keys) if pc else None
+        # Prometheus metric name must match the regex `[a-zA-Z_:][a-zA-Z0-9_:]*`
+        # `.` is a common character in metrics, so here replace it with `:`
+        self._name = self._name.replace(".", ":")
+        self._tag_keys = self._tag_keys + (
+            "host",
+            "pid",
         )
+        self._tags = {"host": socket.gethostname(), "pid": os.getpid()}
+        try:
+            self._metric = (
+                pc.Gauge(self._name, self._description, self._tag_keys) if pc else None
+            )
+        except ValueError:  # pragma: no cover
+            self._metric = None
 
     def _record(self, value=1, tags: Optional[Dict[str, str]] = None):
         if self._metric:
-            if tags:
-                self._metric.labels(**tags).set(value)
+            if tags is not None:
+                tags.update(self._tags)
             else:
-                self._metric.set(value)
+                tags = self._tags
+            self._metric.labels(**tags).set(value)
 
 
 class Counter(PrometheusMetricMixin, AbstractCounter):

mars/metrics/backends/prometheus/tests/test_prometheus_metric.py

@@ -54,7 +54,8 @@ def test_counter(start_prometheus_http_server):
     c = Counter("test_counter", "A test counter", ("service", "tenant"))
     assert c.name == "test_counter"
     assert c.description == "A test counter"
-    assert c.tag_keys == ("service", "tenant")
+    assert set(["host", "pid"]).issubset(set(c.tag_keys))
+    assert set(["service", "tenant"]).issubset(set(c.tag_keys))
     assert c.type == "counter"
     c.record(1, {"service": "mars", "tenant": "test"})
     verify_metric("test_counter", 1.0)
@@ -66,7 +67,7 @@ def test_gauge(start_prometheus_http_server):
     g = Gauge("test_gauge", "A test gauge")
     assert g.name == "test_gauge"
     assert g.description == "A test gauge"
-    assert g.tag_keys == ()
+    assert set(["host", "pid"]).issubset(set(g.tag_keys))
     assert g.type == "gauge"
     g.record(0.1)
     verify_metric("test_gauge", 0.1)
@@ -78,7 +79,7 @@ def test_meter(start_prometheus_http_server):
     m = Meter("test_meter")
     assert m.name == "test_meter"
     assert m.description == ""
-    assert m.tag_keys == ()
+    assert set(["host", "pid"]).issubset(set(m.tag_keys))
     assert m.type == "meter"
     num = 3
     while num > 0:
@@ -92,7 +93,7 @@ def test_histogram(start_prometheus_http_server):
     h = Histogram("test_histogram")
     assert h.name == "test_histogram"
     assert h.description == ""
-    assert h.tag_keys == ()
+    assert set(["host", "pid"]).issubset(set(h.tag_keys))
     assert h.type == "histogram"
     num = 3
     while num > 0: