From a315b1e85d1392626e7592dbe06bb3ff35bb3ee8 Mon Sep 17 00:00:00 2001
From: Dimitri Fontaine <dimitri@citusdata.com>
Date: Mon, 6 Dec 2021 17:50:27 +0100
Subject: [PATCH] Add support for more environment variables. (#846)

* Add support for more environment variables.

The parameters --replication-quorum, --candidate-priority, and --name can
now also be passed from the environment variables
PG_AUTOCTL_REPLICATION_QUORUM, PG_AUTOCTL_CANDIDATE_PRIORITY, and
PG_AUTOCTL_NODE_NAME.

This allows easier Dockerfile "control-at-a-distance" integration style,
where all the containers are the same with the same command, but the
environment is where you can inject container/node/pod specific options.

Co-authored-by: Jelte Fennema <github-tech@jeltef.nl>
---
 src/bin/pg_autoctl/cli_common.c | 54 +++++++++++++++++++++++++++++++++
 src/bin/pg_autoctl/defaults.h   |  5 +++
 tests/test_auth.py              | 13 ++++----
 3 files changed, 66 insertions(+), 6 deletions(-)

diff --git a/src/bin/pg_autoctl/cli_common.c b/src/bin/pg_autoctl/cli_common.c
index be0239972..ead5d1a70 100644
--- a/src/bin/pg_autoctl/cli_common.c
+++ b/src/bin/pg_autoctl/cli_common.c
@@ -112,6 +112,60 @@ cli_common_keeper_getopts(int argc, char **argv,
 	/* default to a "primary" in citus node_role terms */
 	LocalOptionConfig.citusRole = CITUS_ROLE_PRIMARY;
 
+	/* add support for environment variable for some of the options */
+	if (env_exists(PG_AUTOCTL_NODE_NAME))
+	{
+		if (!get_env_copy(PG_AUTOCTL_NODE_NAME,
+						  LocalOptionConfig.name,
+						  sizeof(LocalOptionConfig.name)))
+		{
+			/* errors have already been logged */
+			exit(EXIT_CODE_BAD_ARGS);
+		}
+	}
+
+	if (env_exists(PG_AUTOCTL_CANDIDATE_PRIORITY))
+	{
+		char prio[BUFSIZE] = { 0 };
+
+		if (!get_env_copy(PG_AUTOCTL_CANDIDATE_PRIORITY, prio, sizeof(prio)))
+		{
+			/* errors have already been logged */
+			exit(EXIT_CODE_BAD_ARGS);
+		}
+
+		int candidatePriority = strtol(prio, NULL, 10);
+		if (errno == EINVAL || candidatePriority < 0 || candidatePriority > 100)
+		{
+			log_fatal("PG_AUTOCTL_CANDIDATE_PRIORITY environment variable is not valid."
+					  " Valid values are integers from 0 to 100. ");
+			exit(EXIT_CODE_BAD_ARGS);
+		}
+
+		LocalOptionConfig.pgSetup.settings.candidatePriority = candidatePriority;
+	}
+
+	if (env_exists(PG_AUTOCTL_REPLICATION_QUORUM))
+	{
+		char quorum[BUFSIZE] = { 0 };
+
+		if (!get_env_copy(PG_AUTOCTL_REPLICATION_QUORUM, quorum, sizeof(quorum)))
+		{
+			/* errors have already been logged */
+			exit(EXIT_CODE_BAD_ARGS);
+		}
+
+		bool replicationQuorum = false;
+		if (!parse_bool(quorum, &replicationQuorum))
+		{
+			log_fatal("PG_AUTOCTL_REPLICATION_QUORUM environment variable is not valid."
+					  " Valid values are \"true\" or \"false\".");
+			exit(EXIT_CODE_BAD_ARGS);
+		}
+
+		LocalOptionConfig.pgSetup.settings.replicationQuorum = replicationQuorum;
+	}
+
 	optind = 0;
 
 	while ((c = getopt_long(argc, argv,
diff --git a/src/bin/pg_autoctl/defaults.h b/src/bin/pg_autoctl/defaults.h
index 41b6298cc..edea9ea01 100644
--- a/src/bin/pg_autoctl/defaults.h
+++ b/src/bin/pg_autoctl/defaults.h
@@ -29,6 +29,11 @@
 /* environment variable for --monitor, when used instead of --pgdata */
 #define PG_AUTOCTL_MONITOR "PG_AUTOCTL_MONITOR"
 
+/* environment variable for --candidate-priority and --replication-quorum */
+#define PG_AUTOCTL_NODE_NAME "PG_AUTOCTL_NODE_NAME"
+#define PG_AUTOCTL_CANDIDATE_PRIORITY "PG_AUTOCTL_CANDIDATE_PRIORITY"
+#define PG_AUTOCTL_REPLICATION_QUORUM "PG_AUTOCTL_REPLICATION_QUORUM"
+
 /* default values for the pg_autoctl settings */
 #define POSTGRES_PORT 5432
 #define POSTGRES_DEFAULT_LISTEN_ADDRESSES "*"
diff --git a/tests/test_auth.py b/tests/test_auth.py
index 7cf51e8a7..173d2b8d4 100644
--- a/tests/test_auth.py
+++ b/tests/test_auth.py
@@ -68,12 +68,12 @@ def test_003_init_secondary():
         node1.get_synchronous_standby_names_local(),
         "ANY 1 (pgautofailover_standby_2)",
     )
+
+    # Prevent regression from bug fixed in PR #839
     logs = node2.logs("STDERR")
-    match = re.search(
-        "Failed to connect to .*, retrying until the server is ready", logs
-    )
+    match = re.search("Failed to CHECKPOINT before restart", logs)
     if match:
-        print("Found connection failure in logs: ", match[0])
+        print("Found CHECKPOINT in logs: ", match[0])
     assert not match
     node2.run()
 
@@ -95,8 +95,9 @@ def test_005_logging_of_passwords():
     logs = node2.logs()
     assert monitor_password not in logs
     assert "password=****" in logs
-    # We are still logging passwords when the pguri is incomplete and when printing settings,
-    #  so assert that it's not there in other cases:
+
+    # We are still logging passwords when the pguri is incomplete and when
+    # printing settings, so assert that it's not there in other cases:
     assert not re.search(
         "^(?!primary_conninfo|Failed to find).*%s.*$" % replication_password,
         logs,