(PDB-5672) Analyze reports every hour

postgres (up to at least 16) never analyzes partitioned table parents
Nearly fixed in 14, but removed before release:
  https://www.postgresql.org/about/news/postgresql-14-beta-1-released-2213/
  https://www.postgresql.org/about/news/postgresql-14-rc-1-released-2309/

Assumes the analysis runs quickly enough to avoid needing any enforced
serialization, and to avoid (with the current single threaded
executor) ever delaying gc enough to matter.

src/puppetlabs/puppetdb/cli/services.clj

@@ -45,6 +45,7 @@
             [metrics.timers :refer [time! timer]]
             [metrics.reporters.jmx :as jmx-reporter]
             [murphy :refer [try! with-final]]
+            [next.jdbc :as nxt]
             [puppetlabs.i18n.core :refer [trs tru]]
             [puppetlabs.kitchensink.core :as kitchensink]
             [puppetlabs.puppetdb.cli.tk-util :refer [run-tk-cli-cmd]]
@@ -939,6 +940,18 @@
         (catch InterruptedException _
           (log/info (trs "Garbage collector interrupted")))))))
 
+(defn analyze-partitioned-tables [db shutdown-for-ex]
+  (with-nonfatal-exceptions-suppressed
+    (with-monitored-execution shutdown-for-ex
+      (try!
+        (jdbc/with-transacted-connection db
+          ;; REVIEW: timeout here?
+          (nxt/execute! (jdbc/connection) ["analyze reports"]))
+        (catch InterruptedException _
+          (log/info (trs "reports analysis interrupted")))
+        (catch Exception ex
+          (log/error ex))))))
+
 (defn start-garbage-collection
   "Starts garbage collection of the databases represented in db-configs"
   [{:keys [clean-lock] :as _context}
@@ -952,7 +965,18 @@
           (schedule-with-fixed-delay sched #(invoke-periodic-gc db cfg request
                                                                 shutdown-for-ex
                                                                 clean-lock lock-status)
-                                     (to-millis interval) (to-millis interval)))))))
+                                     (to-millis interval) (to-millis interval))))
+      ;; pg (up to at least 16) never analyzes partitioned table parents
+      ;; Nearly fixed in 14, but removed before release:
+      ;;   https://www.postgresql.org/about/news/postgresql-14-beta-1-released-2213/
+      ;;   https://www.postgresql.org/about/news/postgresql-14-rc-1-released-2309/
+      ;;
+      ;; Assumes the analysis runs quickly enough to avoid needing any
+      ;; enforced serialization, and to avoid (with the current single
+      ;; threaded executor) ever delaying gc enough to matter.
+      (let [hourly (* 60 60 1000)
+            analyze #(analyze-partitioned-tables db shutdown-for-ex)]
+        (schedule-with-fixed-delay sched analyze 0 hourly)))))
 
 
 (defn database-lock-status []

test/puppetlabs/puppetdb/cli/services_test.clj

@@ -1,5 +1,6 @@
 (ns puppetlabs.puppetdb.cli.services-test
   (:require [clojure.set :refer [subset?]]
+            [next.jdbc.plan :refer [select-one!]]
             [puppetlabs.http.client.sync :as pl-http]
             [puppetlabs.puppetdb.cli.util :refer [err-exit-status]]
             [puppetlabs.puppetdb.command.constants :as cmd-consts]
@@ -620,3 +621,23 @@
                                       :db-lock-status db-lock-status})
            (is (= 1 (count (jdbc/query ["SELECT * FROM reports_latest"]))))
            (is (empty? (jdbc/query ["SELECT * FROM reports_historical"])))))))))
+
+(deftest reports-analysis
+  ;; For now, just test for the initial invocation
+  (let [start (now)
+        orig-analyze svcs/analyze-partitioned-tables
+        finished (promise)
+        analyze (fn [& args]
+                  (let [result (apply orig-analyze args)]
+                    (deliver finished true)
+                    result))]
+    (with-redefs [svcs/analyze-partitioned-tables analyze]
+      (svc-utils/with-puppetdb-instance
+        (is (deref finished default-timeout-ms nil))
+        (jdbc/with-db-transaction []
+          (let [r (->> [(str "select last_analyze, last_autoanalyze"
+                             "  from pg_stat_user_tables where relname = 'reports'")]
+                       (select-one! (jdbc/connection) [:last_analyze :last_autoanalyze]))]
+            (is (= nil (:last_autoanalyze r)))
+            (is (time/after? (-> r :last_analyze .getTime time/from-long)
+                             start))))))))