replck or how to fix a replication halt without re-initalization
The replck script utility and its companion low level dsrepair binary allow customers to permanently fix many of the replication halts responding to the following pattern: an infinite loop on CSN when being applied at the consumer level for which the particular CSN is being refused with err=32, err=16 or err=68.
dsrepair is delivered inside the DS product both in the 5.2 series since 5.2patch5 and in the 6.x series. dsrepair is basically split in two parts: 1) a DS plugin that is able to patch the DB contents and 2) a low-level executable which is able to send commands and instructions to that plugin.
replck is a script utility which is currently only delivered via Sun Support channels. Replck would be the equivalent to "fsck" for replication: it inspects the state of replication and, if critical errors are identified on a particular entry, it fixes them by ordering dsrepair to patch the entry contents on the consumer DBs with the entry contents replck extracted from the supplier DB.
Installation:
1) upgrade your DS to 5.2patch5 or 6.x
2) Contact Sun Support to get a copy of the latest available replck delivery
3) Follow the replck README instructions to install replck in your topology: strictely speaking, replck will be required on the supplier machine of the broken link only. It is a high level command tool which can monitor replication and, in case a critical error is found, call the dsrepair low-level tool with the appropriate parameters to patch the DB in the correct way to fix the critical error
forever.
4) Make sure the dsrepair plugin is activated on the consumer machine of the broken link you aim to repair.
5) Run replck like:
$SUPPLIERROOT/shared/bin/replck
and let the tool guide you for the repairing activity. Good to know: replck comes with an extensive documentation in HTML to solve most of the FAQs... It is extremely important to read such documentation in case the replication outage affects more than one level of replicas (i.e., the broken consumer is not a leave in the topology), as in such cases, the way/order in which replck instructions are undertaken will be extremely important.
