1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
|
# Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
namespace=vespa.config.content
## Redundancy decides how many copies of documents will be stored. Documents
## will store copies on enough nodes, such that the sum of the node reliability
## factors is equal or above the redundancy.
redundancy int default=3
## Initial redundancy allows put-operations to return as completed after
## a subset of all copies have been persisted (known as initial persistence
## in BCP terms).
## Set to 0 to disable and use normal redundancy behavior instead.
initial_redundancy int default=0
## If set, requires that the primary nodes copy should always be persisted
## before an n-of-m initial persistence operation can complete.
ensure_primary_persisted bool default=true
## The number of copies that should be "ready" to be active.
## Maximum is redundancy.
ready_copies int default=0
## If this option is set true, the distributor will try to enforce one active copy
## of buckets per leaf hierarchical group. This is a simple implementation for
## search to be able to setup top level dispatcher to only send search to all
## nodes in one group as they have a static cost per node used. It used,
## hierarchical grouping can not be used for other purposes. Using this option
## implies that:
## - ready_copies == redundancy
## - Only one level of hierarchical grouping defined.
## - That level distributes copies to all defined groups.
active_per_leaf_group bool default=false
## Search have some unfortunate properties with some queries, adding a static
## query cost, independent of the number of documents searched on a node. For
## these cases they use many small groups to be able to scale. In such cases,
## where groups are small, having distributors able to take over when none are
## available in a group is useful.
# TODO: Deprecated and unused, remove in Vespa 9
distributor_auto_ownership_transfer_on_whole_group_down bool default=true
## Hierarchical grouping divides the nodes into a tree of groups. Due to config
## liking flat structures. The tree of groups is represented by a single array
## of groups defined here, where index is a string that can have a form like
## "1.2.1" in order to tell where in the tree structure the group is.
group[].index string
## Each group needs to have a name. Makes it easier to operations that they can
## name their groups according to what they want to call them.
group[].name string
## Capacity of the given group
group[].capacity double default=1
## Partitions define how copies are divided among child groups/nodes.
group[].partitions string default=""
## Leaf groups will have a set of nodes within them. Branch groups will have
## none.
group[].nodes[].index int
# Whether this node is retired.
# The system will migrate all data away from retired nodes such that they can
# eventually be removed without partial data loss
group[].nodes[].retired bool default=false
## Which disk distribution to use. From the current choices, we very much
## recommend using MODULO_BID. The only reason to use any of the earlier one is
## during an upgrade where you dont want to mess up the distribution in case you
## want to revert the upgrade.
##
## MODULO old one (4.0)
## MODULO_INDEX with node index in seed
## MODULO_KNUTH with random(node index) in seed
## MODULO_BID using all used bits, except count bits, and random(node index)
## TODO: Deprecated and unused, remove in Vespa 9
disk_distribution enum { MODULO, MODULO_INDEX, MODULO_KNUTH, MODULO_BID } default=MODULO_BID
|
