Knowledge base
cluster-addresscriticalWritten by the HubdirectiveAMQP address of the broker over which nodes announce themselves and tell each other to re-read settings; defaults to the same broker that carries the mail.database-addresscriticalWritten by the HubdirectiveConnection string for the database that holds every runtime setting of the cluster — throttles, DKIM keys, pools, pauses, suppressions, rewrite rules and response patterns.cluster-exchangeimportantWritten by the HubdirectiveName of the fanout exchange over which nodes announce themselves and control messages travel.database-ttlimportantWritten by the HubdirectiveSeconds between re-reads of the settings database; within that window the engine works entirely from memory and never queries the database per message.server-idimportantWritten by the HubdirectiveNumber identifying this node, used to keep generated message identifiers unique across a cluster.www-urlWritten by the HubdirectiveFull URL other cluster members use to link to this node's console; absent from –help but present in the stock config.cluster-timeoutdirectiveHow long a node waits for other nodes to answer its announcement before it proceeds.database-validatedirectiveWhether the engine verifies that the database schema matches the version it expects before it starts working.www-hostdirectiveHostname this node publishes for cluster console links, when a full URL is not given.How a setting reaches a running nodecriticalWritten by the HubmechanismA MailerQ setting lives in the database and reaches a node one of two ways: quietly, when the node's cache expires, or at once, when someone publishes a reload message to the cluster exchange.One database, every node, no mastercriticalWritten by the HubmechanismEvery node reads the same 33 tables and none of them owns the data; a change is a row plus a reload message, and everything else follows from that.Save, reload, and a list that does not refreshimportantWritten by the HubmechanismA console save writes the database and makes the engine re-read at once — but the screen showing the result is stale until reloaded.Make the engine re-read the databasecriticalWritten by the HubprocedureThe engine does not notice database changes on its own; one broker message makes it re-read at once.purge-databasecriticalstartup flagEmpties the cluster database and rebuilds an empty schema — every throttle, DKIM key, pool, pause and suppression is gone.repair-databasecriticalWritten by the Hubstartup flagCreates missing tables and brings an existing schema up to the version the binary expects, then exits.clustercriticalWritten by the HubqueueThe fanout exchange over which nodes announce themselves and control messages travel.amq.gen-*queueThe anonymous queue each node binds to the cluster exchange to receive announcements.