Add monitoring for duply backup

5 files changed, 117 insertions, 5 deletions

diff --git a/modules/private/monitoring/default.nix b/modules/private/monitoring/default.nix
index 8ae0b30..316c2dd 100644
--- a/modules/private/monitoring/default.nix
+++ b/modules/private/monitoring/default.nix
@@ -58,6 +58,9 @@ let
     wrapProgram $out/check_eriomem --prefix PATH : ${lib.makeBinPath [
       pkgs.s3cmd pkgs.python3
     ]}
+    wrapProgram $out/check_eriomem_age --prefix PATH : ${lib.makeBinPath [
+      pkgs.duplicity
+    ]} --set SECRETS_PATH ${lib.optionalString cfg.master config.secrets.fullPaths."eriomem_access_key"}
     wrapProgram $out/notify_by_email --prefix PATH : ${lib.makeBinPath [
       pkgs.mailutils
     ]}
@@ -136,7 +139,7 @@ let
       lib.attrsets.optionalAttrs
         (builtins.pathExists specific_file)
         (pkgs.callPackage specific_file {
-          inherit config emailCheck;
+          inherit config nodes emailCheck;
           hostFQDN = config.hostEnv.fqdn;
           hostName = name;
         });
@@ -232,10 +235,22 @@ in
         dest = "naemon/id_rsa";
         user = "naemon";
         group = "naemon";
-        premissions = "0400";
+        permissions = "0400";
         text = config.myEnv.monitoring.ssh_secret_key;
       }
-    ];
+    ] ++ lib.optional cfg.master (
+      {
+        dest = "eriomem_access_key";
+        user = "naemon";
+        group = "naemon";
+        permissions = "0400";
+        text = ''
+          export AWS_ACCESS_KEY_ID="${config.myEnv.backup.accessKeyId}"
+          export AWS_SECRET_ACCESS_KEY="${config.myEnv.backup.secretAccessKey}"
+          export BASE_URL="${config.myEnv.backup.remote}"
+        '';
+      }
+    );
     # needed since extraResource is not in the closure
     systemd.services.naemon.path = [ myplugins ];
     services.naemon = {
diff --git a/modules/private/monitoring/objects_common.nix b/modules/private/monitoring/objects_common.nix
index 2585c38..c0a17e6 100644
--- a/modules/private/monitoring/objects_common.nix
+++ b/modules/private/monitoring/objects_common.nix
@@ -94,6 +94,7 @@ in
     check_emails = "$USER2$/check_emails -H $HOSTADDRESS$ -i $USER203$ -l $ARG1$ -p $ARG2$ -s $ARG3$ -f $ARG4$";
     check_emails_local = "$USER2$/check_emails -H $HOSTADDRESS$ -n $ARG1$ -r $ADMINEMAIL$ -s $ARG2$ -f $ARG3$";
     check_eriomem = "$USER2$/check_eriomem $USER208$";
+    check_eriomem_age = "$USER2$/check_eriomem_age $ARG1$";
     check_external_dns = "$USER1$/check_dns -H $ARG2$ -s $ARG1$ $ARG3$";
     check_ftp_database = "$USER2$/check_ftp_database";
     check_git = "$USER2$/check_git $USER203$";
diff --git a/modules/private/monitoring/objects_monitoring-1.nix b/modules/private/monitoring/objects_monitoring-1.nix
index ec6fdce..02870ed 100644
--- a/modules/private/monitoring/objects_monitoring-1.nix
+++ b/modules/private/monitoring/objects_monitoring-1.nix
@@ -1,4 +1,23 @@
-{ config, pkgs, hostFQDN, emailCheck, ... }:
+{ config, pkgs, nodes, hostFQDN, emailCheck, ... }:
+let
+  to_eriomem_age_dependency = name: {
+    dependent_host_name = "eldiron.immae.eu";
+    host_name           = "eldiron.immae.eu";
+    dependent_service_description = "Eriomem backup for ${name} is not too old";
+    service_description = "Eriomem backup is up and not full";
+    execution_failure_criteria = "u";
+    notification_failure_criteria = "u";
+  };
+  to_eriomem_age = name: {
+    service_description = "Eriomem backup for ${name} is not too old";
+    host_name = "eldiron.immae.eu";
+    use = "external-service";
+    check_command = ["check_eriomem_age" name];
+
+    check_interval = "120";
+    notification_interval = "120";
+  };
+in
 {
   host = {
     # Dummy host for testing
@@ -654,7 +673,7 @@
       _webstatus_name = "LDAP";
       _webstatus_url = "ldap.immae.eu";
     }
-  ];
+  ] ++ map to_eriomem_age (builtins.attrNames nodes.eldiron.config.services.duplyBackup.profiles);
   contact = {
     telio-tortay = config.myEnv.monitoring.contacts.telio-tortay // {
       use = "generic-contact";
@@ -664,4 +683,5 @@
   contactgroup = {
     telio-tortay = { alias = "Telio Tortay"; members = "immae"; };
   };
+  servicedependency = map to_eriomem_age_dependency (builtins.attrNames nodes.eldiron.config.services.duplyBackup.profiles);
 }
diff --git a/modules/private/monitoring/plugins/check_eriomem_age b/modules/private/monitoring/plugins/check_eriomem_age
new file mode 100755
index 0000000..4d03b82
--- /dev/null
+++ b/modules/private/monitoring/plugins/check_eriomem_age
@@ -0,0 +1,66 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+source $SECRETS_PATH
+export HOME=$(mktemp -d)
+
+trap "rm -rf $HOME" EXIT
+folder=$1
+
+parse_date() {
+  d=$1
+  echo $d | sed -e "s/^\(....\)\(..\)\(..\)T\(..\)\(..\)\(..\)/\1-\2-\3T\4:\5:\6/"
+}
+
+output=$(duplicity collection-status --log-fd 2 "$BASE_URL$folder" 2>&1 > /dev/null)
+
+output=$(echo "$output" | grep -v "^\.")
+
+last_full=$(parse_date $(echo "$output" | grep "^ full " | cut -d' ' -f3 | sort | tail -n1))
+last_bkp=$(parse_date $(echo "$output" | grep -E "^ (full|inc) " | cut -d' ' -f3 | sort | tail -n1))
+orphaned_sets=$(echo "$output" | grep "^orphaned-sets-num" | cut -d' ' -f2)
+incomplete_sets=$(echo "$output" | grep "^incomplete-sets-num" | cut -d' ' -f2)
+
+if [[ -z "$last_full" || -z "$last_bkp" || -z "$orphaned_sets" || -z "$incomplete_sets" ]]; then
+  echo "duply-backup $folder UNKNOWN - impossible to parse result"
+  exit 3
+fi
+
+last_full_age=$(( ($(date "+%s") - $(date -d "$last_full" "+%s")) / (60*60*24) ))
+last_bkp_age=$(( ($(date "+%s") - $(date -d "$last_bkp" "+%s")) / (60*60) ))
+
+PERFS="orphan=$orphaned_sets;1;;0; incomplete=$incomplete_sets;1;;0; age=${last_bkp_age}h;30;48;0; full_age=${last_full_age}d;35;45;0;"
+
+
+WARNINGS=""
+ERRORS=""
+if [[ "$incomplete_sets" -gt 0 ]]; then
+  WARNINGS="$WARNINGS - Incomplete sets is $incomplete_sets"
+fi
+
+if [[ "$orphaned_sets" -gt 0 ]]; then
+  WARNINGS="$WARNINGS - Orphaned sets is $orphaned_sets"
+fi
+
+if [[ "$last_full_age" -gt 45 ]]; then
+  ERRORS="$ERRORS - Last full backup is too old $last_full"
+elif [[ "$last_full_age" -gt 35 ]]; then
+  WARNINGS="$WARNINGS - Last full backup is getting old $last_full"
+fi
+
+if [[ "$last_bkp_age" -gt 48 ]]; then
+  ERRORS="$ERRORS - Last backup is too old $last_bkp"
+elif [[ "$last_bkp_age" -gt 30 ]]; then
+  WARNINGS="$WARNINGS - Last backup is getting old $last_bkp"
+fi
+
+if [[ -n "$ERRORS" ]]; then
+  echo "duply-backup $folder CRITICAL$ERRORS$WARNINGS | $PERFS"
+  exit 2
+elif [[ -n "$WARNINGS" ]]; then
+  echo "duply-backup $folder WARNING$WARNINGS | $PERFS"
+  exit 1
+else
+  echo "duply-backup $folder OK | $PERFS"
+fi
diff --git a/modules/private/monitoring/to_objects.nix b/modules/private/monitoring/to_objects.nix
index 7b4b523..12721d2 100644
--- a/modules/private/monitoring/to_objects.nix
+++ b/modules/private/monitoring/to_objects.nix
@@ -38,7 +38,15 @@ let
     ) v)}
     }
     '';
+  toOtherNoName = keyname: v: ''
+    define ${keyname} {
+    ${builtins.concatStringsSep "\n" (mapAttrsToList (kk: vv:
+      "  ${pad 30 kk}   ${vv}"
+    ) v)}
+    }
+    '';
   toOthers = keyname: a: builtins.concatStringsSep "\n" (mapAttrsToList (toOther keyname) a);
+  toOthersArray = keyname: a: builtins.concatStringsSep "\n" (map (toOtherNoName keyname) a);
 
   toTemplate = keyname: k: v: ''
     define ${keyname} {
@@ -61,6 +69,8 @@ let
       then toTemplates v
     else if builtins.elem keyname ["hostgroup" "host" "contactgroup" "contact" "timeperiod" "servicegroup"]
       then toOthers keyname v
+    else if builtins.elem keyname ["servicedependency"]
+      then toOthersArray keyname v
     else "";
   toObjects = v: builtins.concatStringsSep "\n" (mapAttrsToList toObjects' v);
 in