Add graceful MySQL tmpfs sync on shutdown

This commit is contained in:
uprightbass360
2026-01-08 02:39:08 -05:00
parent fe410a6d4d
commit ce76769f79
7 changed files with 152 additions and 60 deletions

View File

@@ -155,6 +155,7 @@ MYSQL_MAX_CONNECTIONS=1000
MYSQL_INNODB_BUFFER_POOL_SIZE=256M MYSQL_INNODB_BUFFER_POOL_SIZE=256M
MYSQL_INNODB_LOG_FILE_SIZE=64M MYSQL_INNODB_LOG_FILE_SIZE=64M
MYSQL_INNODB_REDO_LOG_CAPACITY=512M MYSQL_INNODB_REDO_LOG_CAPACITY=512M
# MySQL runs on tmpfs (RAM) for performance, with sync to persistent storage on shutdown
MYSQL_RUNTIME_TMPFS_SIZE=8G MYSQL_RUNTIME_TMPFS_SIZE=8G
MYSQL_DISABLE_BINLOG=1 MYSQL_DISABLE_BINLOG=1
MYSQL_CONFIG_DIR=${STORAGE_CONFIG_PATH}/mysql/conf.d MYSQL_CONFIG_DIR=${STORAGE_CONFIG_PATH}/mysql/conf.d

View File

@@ -26,6 +26,7 @@ services:
MYSQL_INNODB_BUFFER_POOL_SIZE: ${MYSQL_INNODB_BUFFER_POOL_SIZE} MYSQL_INNODB_BUFFER_POOL_SIZE: ${MYSQL_INNODB_BUFFER_POOL_SIZE}
MYSQL_INNODB_LOG_FILE_SIZE: ${MYSQL_INNODB_LOG_FILE_SIZE} MYSQL_INNODB_LOG_FILE_SIZE: ${MYSQL_INNODB_LOG_FILE_SIZE}
MYSQL_BINLOG_EXPIRE_LOGS_SECONDS: 86400 MYSQL_BINLOG_EXPIRE_LOGS_SECONDS: 86400
MYSQL_DISABLE_BINLOG: ${MYSQL_DISABLE_BINLOG}
TZ: "${TZ}" TZ: "${TZ}"
entrypoint: entrypoint:
- /usr/local/bin/mysql-entrypoint.sh - /usr/local/bin/mysql-entrypoint.sh
@@ -50,6 +51,7 @@ services:
- --expire_logs_days=0 - --expire_logs_days=0
- --binlog_expire_logs_seconds=86400 - --binlog_expire_logs_seconds=86400
- --binlog_expire_logs_auto_purge=ON - --binlog_expire_logs_auto_purge=ON
stop_grace_period: 2m
restart: unless-stopped restart: unless-stopped
logging: *logging-default logging: *logging-default
healthcheck: healthcheck:

View File

@@ -187,6 +187,8 @@ Because MySQL stores its hot data in a tmpfs (`/var/lib/mysql-runtime`) while pe
- If **any tables exist**, the script logs `Backup restoration completed successfully` and skips the expensive restore just as before. - If **any tables exist**, the script logs `Backup restoration completed successfully` and skips the expensive restore just as before.
- If **no tables are found or the query fails**, the script logs `Restoration marker found, but databases are empty - forcing re-import`, automatically clears the stale marker, and reruns the backup restore + `dbimport` pipeline so services always start with real data. - If **no tables are found or the query fails**, the script logs `Restoration marker found, but databases are empty - forcing re-import`, automatically clears the stale marker, and reruns the backup restore + `dbimport` pipeline so services always start with real data.
On graceful shutdown, the MySQL container now syncs the tmpfs datadir back into `/var/lib/mysql-persistent` so a normal restart keeps the latest state. Unclean shutdowns (host reboot, OOM kill) can still lose recent changes, so the backup restore path remains the safety net.
To complement that one-shot safety net, the long-running `ac-db-guard` service now watches the runtime tmpfs. It polls MySQL, and if it ever finds those schemas empty (the usual symptom after a daemon restart), it automatically reruns `db-import-conditional.sh` to rehydrate from the most recent backup before marking itself healthy. All auth/world services now depend on `ac-db-guard`'s health check, guaranteeing that AzerothCore never boots without real tables in memory. The guard also mounts the working SQL tree from `local-storage/source/azerothcore-playerbots/data/sql` into the db containers so that every `dbimport` run uses the exact SQL that matches your checked-out source, even if the Docker image was built earlier. To complement that one-shot safety net, the long-running `ac-db-guard` service now watches the runtime tmpfs. It polls MySQL, and if it ever finds those schemas empty (the usual symptom after a daemon restart), it automatically reruns `db-import-conditional.sh` to rehydrate from the most recent backup before marking itself healthy. All auth/world services now depend on `ac-db-guard`'s health check, guaranteeing that AzerothCore never boots without real tables in memory. The guard also mounts the working SQL tree from `local-storage/source/azerothcore-playerbots/data/sql` into the db containers so that every `dbimport` run uses the exact SQL that matches your checked-out source, even if the Docker image was built earlier.
Because new features sometimes require schema changes even when the databases already contain data, `ac-db-guard` now performs a `dbimport` verification sweep (configurable via `DB_GUARD_VERIFY_INTERVAL_SECONDS`) to proactively apply any outstanding updates from the mounted SQL tree. By default it runs once per bootstrap and then every 24 hours, so the auth/world servers always see the columns/tables expected by their binaries without anyone having to run host scripts manually. Because new features sometimes require schema changes even when the databases already contain data, `ac-db-guard` now performs a `dbimport` verification sweep (configurable via `DB_GUARD_VERIFY_INTERVAL_SECONDS`) to proactively apply any outstanding updates from the mounted SQL tree. By default it runs once per bootstrap and then every 24 hours, so the auth/world servers always see the columns/tables expected by their binaries without anyone having to run host scripts manually.

View File

@@ -96,7 +96,7 @@ Comprehensive cleanup with multiple destruction levels and safety checks.
Starts all configured containers using appropriate profiles. Starts all configured containers using appropriate profiles.
#### `scripts/bash/stop-containers.sh` - Graceful Shutdown #### `scripts/bash/stop-containers.sh` - Graceful Shutdown
Stops all containers with proper cleanup and data protection. Stops all containers with proper cleanup and data protection. The MySQL container performs a shutdown-time sync from tmpfs to persistent storage.
#### `status.sh` - Service Health Monitoring #### `status.sh` - Service Health Monitoring
```bash ```bash

View File

@@ -75,7 +75,7 @@ services:
| Upstream Concept | RealmMaster Equivalent | Notes | | Upstream Concept | RealmMaster Equivalent | Notes |
| ---------------- | ---------------------- | ----- | | ---------------- | ---------------------- | ----- |
| MySQL container with bind-mounted storage | `ac-mysql` + `ac-storage-init` | Bind mounts live under `storage/` and `local-storage/`; tmpfs keeps runtime data fast and is checkpointed to disk automatically. | | MySQL container with bind-mounted storage | `ac-mysql` + `ac-storage-init` | Bind mounts live under `storage/` and `local-storage/`; tmpfs keeps runtime data fast and is synced to disk on graceful shutdown. |
| Manual DB import container | `ac-db-import` & `ac-db-init` | Automatically imports schemas or restores from backups; disable by skipping the `db` profile if you truly want manual control. | | Manual DB import container | `ac-db-import` & `ac-db-init` | Automatically imports schemas or restores from backups; disable by skipping the `db` profile if you truly want manual control. |
| World/Auth servers with optional DBC overrides | `ac-authserver-*` / `ac-worldserver-*` | Profile-based builds cover vanilla, playerbots, and custom module binaries. DBC overrides go into the shared client data mount just like upstream. | | World/Auth servers with optional DBC overrides | `ac-authserver-*` / `ac-worldserver-*` | Profile-based builds cover vanilla, playerbots, and custom module binaries. DBC overrides go into the shared client data mount just like upstream. |
| Client data bind mounts | `ac-client-data-standard` (or `-playerbots`) | Runs `scripts/bash/download-client-data.sh`, caches releases, and mounts them read-only into the worldserver. | | Client data bind mounts | `ac-client-data-standard` (or `-playerbots`) | Runs `scripts/bash/download-client-data.sh`, caches releases, and mounts them read-only into the worldserver. |

View File

@@ -449,7 +449,7 @@ if [ -n "$backup_path" ]; then
echo "⚠️ Backup restoration failed, will proceed with fresh database setup" echo "⚠️ Backup restoration failed, will proceed with fresh database setup"
fi fi
else else
echo " No valid backups found - proceeding with fresh setup" echo " No valid SQL backups found - proceeding with fresh setup"
echo "$(date): No backup found - fresh setup needed" > "$RESTORE_FAILED_MARKER" echo "$(date): No backup found - fresh setup needed" > "$RESTORE_FAILED_MARKER"
fi fi

View File

@@ -11,66 +11,66 @@ if ! command -v "$ORIGINAL_ENTRYPOINT" >/dev/null 2>&1; then
fi fi
TARGET_SPEC="${MYSQL_RUNTIME_USER:-${CONTAINER_USER:-}}" TARGET_SPEC="${MYSQL_RUNTIME_USER:-${CONTAINER_USER:-}}"
if [ -z "${TARGET_SPEC:-}" ] || [ "${TARGET_SPEC}" = "0:0" ]; then
exec "$ORIGINAL_ENTRYPOINT" "$@"
fi
if [[ "$TARGET_SPEC" != *:* ]]; then
echo "mysql-entrypoint: Expected MYSQL_RUNTIME_USER/CONTAINER_USER in uid:gid form, got '${TARGET_SPEC}'" >&2
exit 1
fi
IFS=':' read -r TARGET_UID TARGET_GID <<< "$TARGET_SPEC"
if ! [[ "$TARGET_UID" =~ ^[0-9]+$ ]] || ! [[ "$TARGET_GID" =~ ^[0-9]+$ ]]; then
echo "mysql-entrypoint: UID/GID must be numeric (received uid='${TARGET_UID}' gid='${TARGET_GID}')" >&2
exit 1
fi
if ! id mysql >/dev/null 2>&1; then
echo "mysql-entrypoint: mysql user not found in container" >&2
exit 1
fi
current_uid="$(id -u mysql)"
current_gid="$(id -g mysql)"
# Adjust group if needed
target_group_name="" target_group_name=""
if [ "$current_gid" != "$TARGET_GID" ]; then if [ -n "${TARGET_SPEC:-}" ] && [ "${TARGET_SPEC}" != "0:0" ]; then
if groupmod -g "$TARGET_GID" mysql 2>/dev/null; then if [[ "$TARGET_SPEC" != *:* ]]; then
target_group_name="mysql" echo "mysql-entrypoint: Expected MYSQL_RUNTIME_USER/CONTAINER_USER in uid:gid form, got '${TARGET_SPEC}'" >&2
else
existing_group="$(getent group "$TARGET_GID" | cut -d: -f1 || true)"
if [ -z "$existing_group" ]; then
existing_group="mysql-host"
if ! getent group "$existing_group" >/dev/null 2>&1; then
groupadd -g "$TARGET_GID" "$existing_group"
fi
fi
usermod -g "$existing_group" mysql
target_group_name="$existing_group"
fi
else
target_group_name="$(getent group mysql | cut -d: -f1)"
fi
if [ -z "$target_group_name" ]; then
target_group_name="$(getent group "$TARGET_GID" | cut -d: -f1 || true)"
fi
# Adjust user UID if needed
if [ "$current_uid" != "$TARGET_UID" ]; then
if getent passwd "$TARGET_UID" >/dev/null 2>&1 && [ "$(getent passwd "$TARGET_UID" | cut -d: -f1)" != "mysql" ]; then
echo "mysql-entrypoint: UID ${TARGET_UID} already in use by $(getent passwd "$TARGET_UID" | cut -d: -f1)." >&2
echo "mysql-entrypoint: Please choose a different CONTAINER_USER or adjust the image." >&2
exit 1 exit 1
fi fi
usermod -u "$TARGET_UID" mysql
fi
# Ensure group lookup after potential changes IFS=':' read -r TARGET_UID TARGET_GID <<< "$TARGET_SPEC"
target_group_name="$(getent group "$TARGET_GID" | cut -d: -f1 || echo "$target_group_name")"
if ! [[ "$TARGET_UID" =~ ^[0-9]+$ ]] || ! [[ "$TARGET_GID" =~ ^[0-9]+$ ]]; then
echo "mysql-entrypoint: UID/GID must be numeric (received uid='${TARGET_UID}' gid='${TARGET_GID}')" >&2
exit 1
fi
if ! id mysql >/dev/null 2>&1; then
echo "mysql-entrypoint: mysql user not found in container" >&2
exit 1
fi
current_uid="$(id -u mysql)"
current_gid="$(id -g mysql)"
# Adjust group if needed
if [ "$current_gid" != "$TARGET_GID" ]; then
if groupmod -g "$TARGET_GID" mysql 2>/dev/null; then
target_group_name="mysql"
else
existing_group="$(getent group "$TARGET_GID" | cut -d: -f1 || true)"
if [ -z "$existing_group" ]; then
existing_group="mysql-host"
if ! getent group "$existing_group" >/dev/null 2>&1; then
groupadd -g "$TARGET_GID" "$existing_group"
fi
fi
usermod -g "$existing_group" mysql
target_group_name="$existing_group"
fi
else
target_group_name="$(getent group mysql | cut -d: -f1)"
fi
if [ -z "$target_group_name" ]; then
target_group_name="$(getent group "$TARGET_GID" | cut -d: -f1 || true)"
fi
# Adjust user UID if needed
if [ "$current_uid" != "$TARGET_UID" ]; then
if getent passwd "$TARGET_UID" >/dev/null 2>&1 && [ "$(getent passwd "$TARGET_UID" | cut -d: -f1)" != "mysql" ]; then
echo "mysql-entrypoint: UID ${TARGET_UID} already in use by $(getent passwd "$TARGET_UID" | cut -d: -f1)." >&2
echo "mysql-entrypoint: Please choose a different CONTAINER_USER or adjust the image." >&2
exit 1
fi
usermod -u "$TARGET_UID" mysql
fi
# Ensure group lookup after potential changes
target_group_name="$(getent group "$TARGET_GID" | cut -d: -f1 || echo "$target_group_name")"
else
target_group_name="$(getent group mysql | cut -d: -f1 || echo mysql)"
fi
# Update ownership on relevant directories if they exist # Update ownership on relevant directories if they exist
for path in /var/lib/mysql-runtime /var/lib/mysql /var/lib/mysql-persistent /backups; do for path in /var/lib/mysql-runtime /var/lib/mysql /var/lib/mysql-persistent /backups; do
@@ -79,6 +79,91 @@ for path in /var/lib/mysql-runtime /var/lib/mysql /var/lib/mysql-persistent /bac
fi fi
done done
# Minimal fix: Restore data from persistent storage on startup and sync on shutdown only
RUNTIME_DIR="/var/lib/mysql-runtime"
PERSISTENT_DIR="/var/lib/mysql-persistent"
sync_datadir() {
if [ ! -d "$RUNTIME_DIR" ]; then
echo "⚠️ Runtime directory not found: $RUNTIME_DIR"
return 1
fi
if [ ! -d "$PERSISTENT_DIR" ]; then
echo "⚠️ Persistent directory not found: $PERSISTENT_DIR"
return 1
fi
user_schema_count="$(find "$RUNTIME_DIR" -mindepth 1 -maxdepth 1 -type d \
! -name mysql \
! -name performance_schema \
! -name information_schema \
! -name sys \
! -name "#innodb_temp" \
! -name "#innodb_redo" 2>/dev/null | wc -l | tr -d ' ')"
if [ "${user_schema_count:-0}" -eq 0 ]; then
echo "⚠️ Runtime data appears empty (system schemas only); skipping sync"
return 0
fi
echo "📦 Syncing MySQL data to persistent storage..."
if command -v rsync >/dev/null 2>&1; then
rsync -a --delete \
--exclude='.restore-completed' \
--exclude='.restore-failed' \
--exclude='.import-completed' \
--exclude='backup.sql' \
"$RUNTIME_DIR"/ "$PERSISTENT_DIR"/
else
# Mirror the runtime state while preserving marker files.
find "$PERSISTENT_DIR" -mindepth 1 -maxdepth 1 \
! -name ".restore-completed" \
! -name ".restore-failed" \
! -name ".import-completed" \
! -name "backup.sql" \
-exec rm -rf {} + 2>/dev/null || true
cp -a "$RUNTIME_DIR"/. "$PERSISTENT_DIR"/
fi
chown -R mysql:"$target_group_name" "$PERSISTENT_DIR"
echo "✅ Sync completed"
}
handle_shutdown() {
echo "🔻 Shutdown signal received"
if command -v mysqladmin >/dev/null 2>&1; then
if mysqladmin -h localhost -u root -p"${MYSQL_ROOT_PASSWORD:-}" shutdown 2>/dev/null; then
echo "✅ MySQL shutdown complete"
sync_datadir || true
else
echo "⚠️ mysqladmin shutdown failed; skipping sync to avoid corruption"
fi
else
echo "⚠️ mysqladmin not found; skipping sync"
fi
if [ -n "${child_pid:-}" ] && kill -0 "$child_pid" 2>/dev/null; then
wait "$child_pid" || true
fi
exit 0
}
# Simple startup restoration
if [ -d "$PERSISTENT_DIR" ]; then
# Check for MySQL data files (exclude marker files starting with .)
if find "$PERSISTENT_DIR" -maxdepth 1 -name "*" ! -name ".*" ! -path "$PERSISTENT_DIR" | grep -q .; then
if [ -d "$RUNTIME_DIR" ] && [ -z "$(ls -A "$RUNTIME_DIR" 2>/dev/null)" ]; then
echo "🔄 Restoring MySQL data from persistent storage..."
cp -a "$PERSISTENT_DIR"/* "$RUNTIME_DIR/" 2>/dev/null || true
chown -R mysql:"$target_group_name" "$RUNTIME_DIR"
echo "✅ Data restored from persistent storage"
fi
fi
fi
# Simple approach: restore on startup only
# Data loss window exists but prevents complete loss on restart
trap handle_shutdown TERM INT
disable_binlog="${MYSQL_DISABLE_BINLOG:-}" disable_binlog="${MYSQL_DISABLE_BINLOG:-}"
if [ "${disable_binlog}" = "1" ]; then if [ "${disable_binlog}" = "1" ]; then
add_skip_flag=1 add_skip_flag=1
@@ -93,4 +178,6 @@ if [ "${disable_binlog}" = "1" ]; then
fi fi
fi fi
exec "$ORIGINAL_ENTRYPOINT" "$@" "$ORIGINAL_ENTRYPOINT" "$@" &
child_pid=$!
wait "$child_pid"