diff options
Diffstat (limited to 'configdefinitions/src/vespa/vespa.config.content.persistence.def')
| -rw-r--r-- | configdefinitions/src/vespa/vespa.config.content.persistence.def | 60 |
1 files changed, 60 insertions, 0 deletions
diff --git a/configdefinitions/src/vespa/vespa.config.content.persistence.def b/configdefinitions/src/vespa/vespa.config.content.persistence.def new file mode 100644 index 00000000000..b2709dedc57 --- /dev/null +++ b/configdefinitions/src/vespa/vespa.config.content.persistence.def @@ -0,0 +1,60 @@ +# Copyright 2017 Yahoo Holdings. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root. +namespace=vespa.config.content + +## If set, causes the service layer to trigger an abort of any queued +## operations for buckets whose ownership changes during a cluster state +## change. This helps ensure consistency between the distributors and +## the storage nodes by discarding operations that were not sent by the +## new owning distributor. The service layer will also block any new +## operations from happening until all operations currently executing for +## buckets with changed ownership have completed. +## Enabling this option has a potential of causing a minor performance +## impact when a node goes up or down due to its blocking semantics, but +## this should be on the order of tens to a few hundred milliseconds. +abort_operations_with_changed_bucket_ownership bool default=true + +## If set, aborts any mutating ideal state operations that arrive to a node +## that were scheduled and sent by distributors that no longer own the buckets +## the operations are bound for. This option complements the above config, +## as this is checked for every arriving message and not when the state changes. +abort_outdated_mutating_ideal_state_ops bool default=true + +## If set, aborts any external (client) operations that arrive to a node from +## a distributor that is no longer the owner of the bucket those operations +## belong to. As with the abort_outdated_mutating_ideal_state_ops option, +## this is a check that happens whenever a message arrives, not when the state +## changes. +abort_outdated_mutating_external_load_ops bool default=true + +## Whether or not to disable partitions when they report I/O errors +fail_partition_on_error bool default=true + +## Time period to keep all updates (given in seconds). One can revert any +## operation done within this time. +revert_time_period int default=300 + +## Minimum number of seconds that document remove (tombstone) entries are +## retained in a bucket before being automatically compacted away. In the case +## where a node has been down for a longer period of time than this value +## and then comes back up, there exists a possibility for documents to be +## resurrected. This happens when documents existing on both the downed node +## and the online nodes are marked as removed on the online nodes and then +## subsequently compacted away along with their remove entries. When the old +## node comes back online, it will appear to the system as if the node has +## documents that do not exist on other nodes (as no remove-entries can be +## found for them) and they will subsequently re-appear. +keep_remove_time_period int default=604800 + +## For providers supporting multiple document versions, this option specifies +## a soft limit on the maximum number of versions that can be stored for a +## single document at any given time. This is considered a soft limit as it +## is possible for more versions to exist concurrently than what this config +## value specifies in a case where the pressure on the containing bucket is not +## yet high enough to warrant a compaction. +## After a compaction has taken place, only the n _newest_ document versions +## will be present in the compacted representation of the bucket. +## This limitation takes precedence over what versions may be stored according +## to revert_time_period. +## Configured value must be at least 1. +maximum_versions_of_single_document_stored int default=5 + |