Cert Failures & BF Aborts (rate/min) ?Certification failures occur when two nodes try to write to the same row simultaneously. BF (Brute Force) aborts happen when a local transaction is rolled back because a remote replicated write took priority. High rates indicate write contention — common during concurrent add-to-cart or order operations.
Flow Control Paused % ?Percentage of time the cluster was paused because a node couldn't keep up with replication. Values above 0.01 (1%) mean writes are being throttled. This is the single most important Galera health metric — any sustained pause directly adds latency to every write.
InnoDB Lock Waits (rate/min) & Current ?Lock Waits/min shows how often transactions are waiting to acquire a row lock. Current Waits shows how many are waiting right now. Spikes correlate with slow transactions — if add-to-cart is slow, you'll likely see lock waits spike here from concurrent cart/inventory updates.
Deadlocks (cumulative) ?Total deadlocks since server start. A deadlock happens when two transactions each hold a lock the other needs. Any increase means concurrent operations are conflicting — often from cart operations hitting the same product/inventory rows.
Galera Queue Avg (recv/send) ?Average length of the receive and send replication queues. Recv queue > 0.5 means this node is falling behind applying remote writes. Send queue > 0.5 means outgoing replication is backed up. Both add latency to write operations.
Galera Commits (rate/min) ?Number of transactions committed per minute across the cluster as seen by this node. This is your write throughput. Correlate with traffic patterns — spikes during peak hours should be matched by stable cert failure and lock wait rates.
Redis Ops/sec ?Commands processed per second on each Redis/KeyDB instance. SG KeyDB handles cache lookups, distributed locks, and Symfony Messenger queue. App Redis handles sessions, cart storage, and framework cache. Sudden drops may indicate connection issues.
Redis Memory (MB) ?Memory usage of each Redis instance. Watch for steady growth that approaches maxmemory (SG KeyDB: 3GB, App Redis: 2GB). When maxmemory is hit, eviction kicks in based on the configured policy, which can cause cache misses and increase DB load.
Cart Keys (App Server db1) ?Number of active cart entries stored in Redis db1. Each key represents an active shopping cart. Correlate with add-to-cart transaction times — more active carts means more concurrent writes to cart-related DB tables (cart, cart_line_item, etc.).
Redis Connected Clients ?Number of active client connections to each Redis instance. Sudden spikes may indicate connection pool exhaustion or a connection leak. Each Shopware worker maintains persistent connections, so this should be relatively stable.
Galera Cluster
Cluster Health ?Overall Galera cluster state. Primary = cluster has quorum. Size = number of active nodes (3 = healthy with arbitrator). Synced = node is fully caught up with the cluster.
Flow Control ?Flow control throttles writes when a node falls behind. FC Paused % > 0 means writes are being delayed. FC Recv/Sent count flow control messages. Queue averages show replication backlog depth.
Certification ?Galera uses certification-based replication. Each write is certified against concurrent writes on other nodes. Failures mean conflicting writes were detected and one was rolled back. High cert failures + BF aborts = hot row contention.
InnoDB Locks ?Row-level locking metrics from InnoDB. Current Waits = transactions blocked right now. Lock Time Avg = average wait time in ms. High averages with concurrent add-to-cart suggest contention on cart/product/inventory tables. Deadlocks require transaction retry.
Processes Waiting on Locks ?Live view of database queries currently blocked waiting for a lock. Shows which queries are stuck and for how long. Long-running entries here are actively causing user-visible latency.