Implement multithreading for database checks with pgbackrest_auto (#18)

vitabaks · Nov 30, 2023 · f0b9307 · f0b9307
1 parent 4e41f61
commit f0b9307
Show file tree

Hide file tree

Showing 2 changed files with 93 additions and 49 deletions.
diff --git a/README.md b/README.md
@@ -78,14 +78,14 @@ Support three types of restore:
 
 Important: Run on the nodes on which you want to restore the backup
 
-Usage: /usr/bin/pgbackrest_auto --from=STANZANAME --to=DATA_DIRECTORY [ --datname=DATABASE [...] ] [ --recovery-type=( default | immediate | time ) ] [ --recovery-target=TIMELINE  [ --backup-set=SET ] [ --backup-host=HOST ] [ --pgver= ] [ --checkdb ] [ --clear ] [ --report ] ]
+Usage: /usr/bin/pgbackrest_auto --from=STANZANAME --to=DATA_DIRECTORY [ --datname=DATABASE [...] ] [ --recovery-type=( default | immediate | time ) ] [ --recovery-target=TIMELINE  [ --backup-set=SET ] [ --pgver= ] [ --checkdb ] [ --clear ] [ --report ] ]
 
 --from=STANZANAME
         Stanza from which you need to restore from a backup
 
 --to=DATA_DIRECTORY
         PostgreSQL Data directory Path to restore from a backup
-        a PostgreSQL database cluster (PGDATA) will be automatically created (initdb) if it does not exist
+        a PostgreSQL database cluster (PGDATA) will be automatically created if it does not exist
         Example: /bkpdata/rst/app-db
 
 --datname=DATABASE [...]
@@ -112,16 +112,12 @@ Usage: /usr/bin/pgbackrest_auto --from=STANZANAME --to=DATA_DIRECTORY [ --datnam
                     incr backup: 20220611-000004F_20220614-000003D
         This is the name of SET: 20220611-000004F_20220614-000003D
 
---backup-host=HOST
-        pgBacRest repository ip address (Use SSH Key-Based Authentication)
-        localhost [default]
-
 --pgver=VERSION
         PostgreSQL cluster (instance) version [ optional ]
         by default, the PostgreSQL version will be determined from the pgbackrest info
 
 --dummy-dump
-        Verify that data can be read out. Check with pg_dump >> /dev/null
+        Verify that data can be read out. Check with pg_dump
 
 --checksums
         Check data checksums
@@ -144,15 +140,17 @@ Usage: /usr/bin/pgbackrest_auto --from=STANZANAME --to=DATA_DIRECTORY [ --datnam
 --config=/path/to/pgbackrest.conf
         The path to the custom pgbackrest configuration file [ optional ]
 
---custom-options=""
+--custom-options=
         Costom options for pgBackRest [ optional ]
-	This includes all the options that may also be configured in pgbackrest.conf
-        Example: "--option1=value --option2=value --option3=value"
+        This includes all the options that may also be configured in pgbackrest.conf
+        Example: --option1=value --option2=value --option3=value
         See all available options: https://pgbackrest.org/configuration.html
 
+--process-max=
+        Max processes to use for restore and validate (default 1).
 
 EXAMPLES:
-( example stanza "app-db" , backup-host "localhost" )
+( example stanza "app-db" , backup host "localhost" (default value) )
 
 | Restore last backup:
 

diff --git a/pgbackrest_auto b/pgbackrest_auto
@@ -8,7 +8,7 @@
 # for "--report": sendemail
 # Run as user: postgres
 
-ver="1.5.1"
+ver="1.6.0"
 
 # variables for function "sendmail()"
 smtp_server="10.128.64.5:25"
@@ -106,6 +106,9 @@ while getopts ":-:" optchar; do
         custom-options=* )
             CUSTOMOTIONS=${OPTARG#*=}
             ;;
+        process-max=* )
+            PROCESS_MAX=${OPTARG#*=}
+            ;;
     esac
 done
 
@@ -160,7 +163,7 @@ Usage: $0 --from=STANZANAME --to=DATA_DIRECTORY [ --datname=DATABASE [...] ] [ -
         by default, the PostgreSQL version will be determined from the pgbackrest info
 
 --dummy-dump
-        Verify that data can be read out. Check with pg_dump >> /dev/null
+        Verify that data can be read out. Check with pg_dump
 
 --checksums
         Check data checksums
@@ -189,6 +192,8 @@ Usage: $0 --from=STANZANAME --to=DATA_DIRECTORY [ --datname=DATABASE [...] ] [ -
         Example: "--option1=value --option2=value --option3=value"
         See all available options: https://pgbackrest.org/configuration.html
 
+--process-max=
+        Max processes to use for restore and validate (default 1).
 
 EXAMPLES:
 ( example stanza \"app-db\" , backup host \"localhost\" (default value) )
@@ -302,15 +307,6 @@ if ! command -v "${PG_BIN_DIR}"/pg_ctl &> /dev/null; then
     exit
 fi
 
-# check if pg_checksums exists (for PostgreSQL version <= 11)
-if [[ "$PGVER" -le "11" && "${CHECKSUMS}" = "yes" ]] || [[ "$PGVER" -le "11" && "${CHECKDB}" = "yes" ]]; then
-    if ! command -v "${PG_BIN_DIR}"/pg_checksums &> /dev/null
-    then
-        warnmsg "pg_checksums command not be found. Please install the postgresql-$PGVER-pg-checksums package"
-        exit
-    fi
-fi
-
 # check if a directory exists
 if [[ ! -d "${PGDATA}" ]]; then
     if ! mkdir -p "${PGDATA}"; then
@@ -354,6 +350,10 @@ elif [[ -n $DATNAME ]]; then
     restore_type_msg="Partial PostgreSQL Restore"
 fi
 
+# process-max default
+if [[ -z $PROCESS_MAX  ]]; then
+    PROCESS_MAX=1
+fi
 
 function sigterm_handler(){
     info "Recieved QUIT|TERM|INT signal"
@@ -513,8 +513,8 @@ function pgbackrest_exec(){
     if [ -f "${detail_rst_log}" ]; then info "See detailed log in the file ${detail_rst_log}"; fi
     info "Restore from backup started. Type: $restore_type_msg"
     # execute pgbackrest
-    echo "pgbackrest --config=${pgbackrest_conf} --stanza=${FROM} --pg1-path=${TO} ${pgbackrest_opt} --delta restore --process-max=4 --log-level-console=error --log-level-file=detail --recovery-option=${recovery_opt} --tablespace-map-all=${TO}_remapped_tablespaces"
-    if bash -c "pgbackrest --config=${pgbackrest_conf} --stanza=${FROM} --pg1-path=${TO} ${pgbackrest_opt} --delta restore --process-max=4 --log-level-console=error --log-level-file=detail --recovery-option=${recovery_opt} --tablespace-map-all=${TO}_remapped_tablespaces"
+    echo "pgbackrest --config=${pgbackrest_conf} --stanza=${FROM} --pg1-path=${TO} ${pgbackrest_opt} --delta restore --process-max=${PROCESS_MAX} --log-level-console=error --log-level-file=detail --recovery-option=${recovery_opt} --tablespace-map-all=${TO}_remapped_tablespaces"
+    if bash -c "pgbackrest --config=${pgbackrest_conf} --stanza=${FROM} --pg1-path=${TO} ${pgbackrest_opt} --delta restore --process-max=${PROCESS_MAX} --log-level-console=error --log-level-file=detail --recovery-option=${recovery_opt} --tablespace-map-all=${TO}_remapped_tablespaces"
     then
         info "Restore from backup done"
         sed -i 's/Restore_from_backup=0/Restore_from_backup=1/g' "${status_file}"
@@ -563,31 +563,68 @@ function dummy_dump(){
         for db in $databases; do
             info "Start data validation for database $db"
             if pgisready 1> /dev/null; then
-                info "  starting pg_dump -p ${PGPORT} -h 127.0.0.1 -d $db >> /dev/null"
-                if ! "${PG_BIN_DIR}"/pg_dump -p "${PGPORT}" -h 127.0.0.1 -d "$db" >> /dev/null
-                then
-                    sed -i 's/Data_validation=1/Data_validation=0/g' "${status_file}"
-                    error "Data validation in the database $db - Failed"
+                if [[ "$PROCESS_MAX" == '1' ]]; then
+                    # single-threaded dump is slower, but requires less disk I/O due to redirection to /dev/null
+                    info "  starting pg_dump -p ${PGPORT} -h 127.0.0.1 -d $db >> /dev/null"
+                    if ! "${PG_BIN_DIR}"/pg_dump -p "${PGPORT}" -h 127.0.0.1 -d "$db" >> /dev/null
+                    then
+                        sed -i 's/Data_validation=1/Data_validation=0/g' "${status_file}"
+                        error "Data validation in the database $db - Failed"
+                    else
+                        info "Data validation in the database $db - Successful"
+                    fi
                 else
-                    info "Data validation in the database $db - Successful"
+                    # Parallel dump speeds up the process for large databases,
+                    # but requires writing data and more disk space for the dump.
+                    mkdir -p "${PGDATA}"/dump/
+                    rm -rf "${PGDATA}"/dump/"$db"
+                    info "  starting pg_dump -p ${PGPORT} -h 127.0.0.1 -d $db -F d -j ${PROCESS_MAX} -f ${PGDATA}/dump/$db"
+                    if ! "${PG_BIN_DIR}"/pg_dump -p "${PGPORT}" -h 127.0.0.1 -d "$db" -F d -j "${PROCESS_MAX}" -f "${PGDATA}"/dump/"$db"
+                    then
+                        sed -i 's/Data_validation=1/Data_validation=0/g' "${status_file}"
+                        error "Data validation in the database $db - Failed"
+                    else
+                        info "Data validation in the database $db - Successful"
+                    fi
                 fi
             fi
         done
 }
 
 # checksums - check data checksums
 function pg_checksums(){
-    if pgisready 1> /dev/null; then pg_stop cycle_simple pg_stop_check; fi
-    info "pg_checksums: starting data checksums validation"
-    sed -i 's/PG_checksums_validation=0/PG_checksums_validation=1/g' "${status_file}"
-    pg_checksums_result=$("${PG_BIN_DIR}"/pg_checksums -c -D "${PGDATA}" | grep "Bad checksums")
-    if [[ $pg_checksums_result != "Bad checksums:  0" ]]
-    then
-        warnmsg "pg_checksums: data checksums validation result: $pg_checksums_result"
-        sed -i 's/PG_checksums_validation=1/PG_checksums_validation=0/g' "${status_file}"
-        error "pg_checksums: data checksums validation - Failed"
+    local pg_checksums_command
+    local pg_checksums_result
+
+    # Determine the checksums command based on PostgreSQL version
+    if [[ "$PGVER" -le "11" ]]; then
+        if command -v "${PG_BIN_DIR}"/pg_checksums &> /dev/null; then
+            pg_checksums_command="pg_checksums"
+        elif command -v "${PG_BIN_DIR}"/pg_verify_checksums &> /dev/null; then
+            pg_checksums_command="pg_verify_checksums"
+        else
+            warnmsg "Checksum command not found. Please install the postgresql-$PGVER-pg-checksums package."
+            exit 1
+        fi
     else
-        info "pg_checksums: data checksums validation - Successful"
+        pg_checksums_command="pg_checksums"
+    fi
+
+    if pgisready 1> /dev/null; then pg_stop cycle_simple pg_stop_check; fi
+        info "pg_checksums: starting data checksums validation"
+        sed -i 's/PG_checksums_validation=0/PG_checksums_validation=1/g' "${status_file}"
+        if [ "$pg_checksums_command" == "pg_verify_checksums" ]; then
+            pg_checksums_result=$("${PG_BIN_DIR}"/pg_verify_checksums -D "${PGDATA}" | grep "Bad checksums")
+        else
+            pg_checksums_result=$("${PG_BIN_DIR}"/pg_checksums -c -D "${PGDATA}" | grep "Bad checksums")
+        fi
+        if [[ $pg_checksums_result != "Bad checksums:  0" ]]
+        then
+            warnmsg "pg_checksums: data checksums validation result: $pg_checksums_result"
+            sed -i 's/PG_checksums_validation=1/PG_checksums_validation=0/g' "${status_file}"
+            error "pg_checksums: data checksums validation - Failed"
+        else
+            info "pg_checksums: data checksums validation - Successful"
     fi
 }
 
@@ -598,7 +635,7 @@ function amcheck_exists(){
     else
         extension='amcheck'
     fi
-    if ! psql -v "ON_ERROR_STOP" -p "${PGPORT}" -h 127.0.0.1 -U postgres -d "$db_name" -tAXc "CREATE EXTENSION if not exists $extension" &> /dev/null
+    if ! psql -v "ON_ERROR_STOP" -p "${PGPORT}" -h 127.0.0.1 -U postgres -d "$db" -tAXc "CREATE EXTENSION if not exists $extension" &> /dev/null
     then
         error "CREATE EXTENSION $extension failed"
     fi
@@ -609,19 +646,28 @@ function amcheck(){
     if ! pgisready 1> /dev/null; then pg_start cycle_simple pgisready; fi
     sed -i 's/Amcheck_validation=0/Amcheck_validation=1/g' "${status_file}"
     databases=$(bash -c "psql -p ${PGPORT} -h 127.0.0.1 -tAXc \"select datname from pg_database where not datistemplate\"")
-    for db_name in $databases; do
+    for db in $databases; do
         if pgisready 1> /dev/null; then
             if amcheck_exists; then
-                info "amcheck: verify the logical consistency of the structure of indexes and heap relations in the database $db_name"
-                indexes=$(psql -p "${PGPORT}" -h 127.0.0.1 -d "$db_name" -tXAc "SELECT quote_ident(n.nspname)||'.'||quote_ident(c.relname) FROM pg_index i JOIN pg_opclass op ON i.indclass[0] = op.oid JOIN pg_am am ON op.opcmethod = am.oid JOIN pg_class c ON i.indexrelid = c.oid JOIN pg_namespace n ON c.relnamespace = n.oid WHERE am.amname = 'btree' AND n.nspname NOT IN ('pg_catalog', 'pg_toast') AND c.relpersistence != 't' AND c.relkind = 'i' AND i.indisready AND i.indisvalid")
-                for index in $indexes; do
-                    # info "amcheck: verify the logical consistency of the structure of index ${index}"
-                    if ! psql -v ON_ERROR_STOP=on -p "${PGPORT}" -h 127.0.0.1 -d "$db_name" -tAXc "select bt_index_parent_check('${index}', heapallindexed => true)" 1> /dev/null
+                info "amcheck: verify the logical consistency of the structure of indexes and heap relations in the database $db"
+                if [[ "$PGVER" -lt '14' ]]; then
+                    # If the PostgreSQL version is less than 14, use the bt_index_parent_check function for each index in single-threaded mode.
+                    indexes=$(psql -p "${PGPORT}" -h 127.0.0.1 -d "$db" -tXAc "SELECT quote_ident(n.nspname)||'.'||quote_ident(c.relname) FROM pg_index i JOIN pg_opclass op ON i.indclass[0] = op.oid JOIN pg_am am ON op.opcmethod = am.oid JOIN pg_class c ON i.indexrelid = c.oid JOIN pg_namespace n ON c.relnamespace = n.oid WHERE am.amname = 'btree' AND n.nspname NOT IN ('pg_catalog', 'pg_toast') AND c.relpersistence != 't' AND c.relkind = 'i' AND i.indisready AND i.indisvalid")
+                    for index in $indexes; do
+                        if ! psql -v ON_ERROR_STOP=on -p "${PGPORT}" -h 127.0.0.1 -d "$db" -tAXc "select bt_index_parent_check('${index}', heapallindexed => true)" 1> /dev/null
+                        then
+                            sed -i 's/Amcheck_validation=1/Amcheck_validation=0/g' "${status_file}"
+                            error "amcheck: logical validation for index ${index} ( database $db ) - Failed"
+                        fi
+                    done
+                else
+                    # Use pg_amcheck (added in PostgreSQL 14)
+                    if ! "${PG_BIN_DIR}"/pg_amcheck -p "${PGPORT}" -h 127.0.0.1 -d "$db" --parent-check --heapallindexed --progress --jobs="${PROCESS_MAX}"
                     then
-                        warnmsg "amcheck: logical validation for index ${index} ( database $db_name ) - Failed"
                         sed -i 's/Amcheck_validation=1/Amcheck_validation=0/g' "${status_file}"
+                        error "amcheck: logical validation for database $db - Failed"
                     fi
-                done
+                fi
             fi
         fi
     done