PostgresNode::pid uses the data from "pg_ctl status" output.

postgrespro · dmitry-lipetsk · Mar 1, 2025 · Feb 27, 2025 · Feb 27, 2025 · Feb 28, 2025
commit 0402c4a6145820043bf94c879921d244ddd8d09a
diff --git a/testgres/consts.py b/testgres/consts.py
@@ -35,3 +35,7 @@
 
 # logical replication settings
 LOGICAL_REPL_MAX_CATCHUP_ATTEMPTS = 60
+
+PG_CTL__STATUS__OK = 0
+PG_CTL__STATUS__NODE_IS_STOPPED = 3
+PG_CTL__STATUS__BAD_DATADIR = 4
diff --git a/testgres/node.py b/testgres/node.py
@@ -49,7 +49,9 @@
     RECOVERY_CONF_FILE, \
     PG_LOG_FILE, \
     UTILS_LOG_FILE, \
-    PG_PID_FILE
+    PG_CTL__STATUS__OK, \
+    PG_CTL__STATUS__NODE_IS_STOPPED, \
+    PG_CTL__STATUS__BAD_DATADIR \
 
 from .consts import \
     MAX_LOGICAL_REPLICATION_WORKERS, \
@@ -132,9 +134,6 @@ class PostgresNode(object):
     # a max number of node start attempts
     _C_MAX_START_ATEMPTS = 5
 
-    # a max number of read pid file attempts
-    _C_MAX_GET_PID_ATEMPTS = 5
-
     def __init__(self, name=None, base_dir=None, port=None, conn_params: ConnectionParams = ConnectionParams(),
                  bin_dir=None, prefix=None):
         """
@@ -211,40 +210,136 @@ def pid(self):
         Return postmaster's PID if node is running, else 0.
         """
 
-        nAttempt = 0
-        pid_file = os.path.join(self.data_dir, PG_PID_FILE)
-        pid_s: str = None
+        self__data_dir = self.data_dir
+
+        _params = [
+            self._get_bin_path('pg_ctl'),
+            "-D", self__data_dir,
+            "status"
+        ]  # yapf: disable
+
+        status_code, out, error = execute_utility2(
+            self.os_ops,
+            _params,
+            self.utils_log_file,
+            verbose=True,
+            ignore_errors=True)
+
+        assert type(status_code) == int  # noqa: E721
+        assert type(out) == str  # noqa: E721
+        assert type(error) == str  # noqa: E721
+
+        # -----------------
+        if status_code == PG_CTL__STATUS__NODE_IS_STOPPED:
+            return 0
+
+        # -----------------
+        if status_code == PG_CTL__STATUS__BAD_DATADIR:
+            return 0
+
+        # -----------------
+        if status_code != PG_CTL__STATUS__OK:
+            errMsg = "Getting of a node status [data_dir is {0}] failed.".format(self__data_dir)
+
+            raise ExecUtilException(
+                message=errMsg,
+                command=_params,
+                exit_code=status_code,
+                out=out,
+                error=error,
+            )
+
+        # -----------------
+        assert status_code == PG_CTL__STATUS__OK
+
+        if out == "":
+            __class__._throw_error__pg_ctl_returns_an_empty_string(
+                _params
+            )
+
+        C_PID_PREFIX = "(PID: "
+
+        i = out.find(C_PID_PREFIX)
+
+        if i == -1:
+            __class__._throw_error__pg_ctl_returns_an_unexpected_string(
+                out,
+                _params
+            )
+
+        assert i > 0
+        assert i < len(out)
+        assert len(C_PID_PREFIX) <= len(out)
+        assert i <= len(out) - len(C_PID_PREFIX)
+
+        i += len(C_PID_PREFIX)
+        start_pid_s = i
+
         while True:
-            if nAttempt == __class__._C_MAX_GET_PID_ATEMPTS:
-                errMsg = "Can't read postmaster pid file [{0}].".format(pid_file)
-                raise Exception(errMsg)
+            if i == len(out):
+                __class__._throw_error__pg_ctl_returns_an_unexpected_string(
+                    out,
+                    _params
+                )
 
-            nAttempt += 1
+            ch = out[i]
 
-            s1 = self.status()
-            if s1 != NodeStatus.Running:
-                return 0
+            if ch == ")":
+                break
 
-            try:
-                lines = self.os_ops.readlines(pid_file)
-            except Exception:
-                s2 = self.status()
-                if s2 == NodeStatus.Running:
-                    raise
-                return 0
-
-            assert lines is not None  # [2025-02-27] OK?
-            assert type(lines) == list  # noqa: E721
-            if len(lines) == 0:
+            if ch.isdigit():
+                i += 1
                 continue
 
-            pid_s = lines[0]
-            assert type(pid_s) == str  # noqa: E721
-            if len(pid_s) == 0:
-                continue
+            __class__._throw_error__pg_ctl_returns_an_unexpected_string(
+                out,
+                _params
+            )
+            assert False
+
+        if i == start_pid_s:
+            __class__._throw_error__pg_ctl_returns_an_unexpected_string(
+                out,
+                _params
+            )
 
-            pid = int(pid_s)
-            return pid
+        # TODO: Let's verify a length of pid string.
+
+        pid = int(out[start_pid_s:i])
+
+        if pid == 0:
+            __class__._throw_error__pg_ctl_returns_a_zero_pid(
+                out,
+                _params
+            )
+
+        assert pid != 0
+        return pid
+
+    @staticmethod
+    def _throw_error__pg_ctl_returns_an_empty_string(_params):
+        errLines = []
+        errLines.append("Utility pg_ctl returns empty string.")
+        errLines.append("Command line is {0}".format(_params))
+        raise RuntimeError("\n".join(errLines))
+
+    @staticmethod
+    def _throw_error__pg_ctl_returns_an_unexpected_string(out, _params):
+        errLines = []
+        errLines.append("Utility pg_ctl returns an unexpected string:")
+        errLines.append(out)
+        errLines.append("------------")
+        errLines.append("Command line is {0}".format(_params))
+        raise RuntimeError("\n".join(errLines))
+
+    @staticmethod
+    def _throw_error__pg_ctl_returns_a_zero_pid(out, _params):
+        errLines = []
+        errLines.append("Utility pg_ctl returns a zero pid. Output string is:")
+        errLines.append(out)
+        errLines.append("------------")
+        errLines.append("Command line is {0}".format(_params))
+        raise RuntimeError("\n".join(errLines))
 
     @property
     def auxiliary_pids(self):