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
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
|
# Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
namespace=vespa.config.content.core
## Root directory for all files related to this storage node.
## Will typically be "$VESPA_HOME/var/db/vespa/vds/<cluster>/<nodetype>/<index>
root_folder string restart
## VDS cluster
cluster_name string default="storage" restart
## The index of this node. Each node of the same type in the same cluster need
## to have unique indexes. This should not be changed, as this is what we use
## to identify the node, and to decide what data should be on it.
node_index int default=0 restart
## Set whether this is a distributor or a storage node. This will decide what
## storage links are set up.
is_distributor bool restart
## Capacity of the node. How much data and load this node will get relative to
## other nodes.
node_capacity double default=1.0 restart
## Capacity of the disks on this node. How much data and load will each disk
## get relative to the other disks on this node.
disk_capacity[] double restart
## Reliability of this node. How much of the cluster redundancy factor can this
## node make up for.
node_reliability int default=1 restart
## The upper bound of merges that any storage node can have active.
## A merge operation will be chained through all nodes involved in the
## merge, only actually starting the operation when every node has
## allowed it to pass through.
## NOTE: these config values are _not_ used if merge_throttling_policy.type
## is configured to DYNAMIC (see below).
max_merges_per_node int default=16
max_merge_queue_size int default=100
## Chooses the throttling policy used to control the active merge window size
## of the MergeThrottler component.
merge_throttling_policy.type enum { STATIC, DYNAMIC } default=STATIC
## Only used if merge_throttling_policy.type == DYNAMIC:
merge_throttling_policy.min_window_size int default=16
merge_throttling_policy.max_window_size int default=128
merge_throttling_policy.window_size_increment double default=2.0
## If the persistence provider indicates that it has exhausted one or more
## of its internal resources during a mutating operation, new merges will
## be bounced for this duration. Not allowing further merges helps take
## load off the node while it e.g. compacts its data stores or memory in
## the background.
## Note: this does not affect merges where the current node is marked as
## "source only", as merges do not cause mutations on such nodes.
resource_exhaustion_merge_back_pressure_duration_secs double default=30.0
## If true, received merges that have already been accepted into the pending
## merge window on at least one node will not be restricted by the configured
## max_merge_queue_size limit. They will be allowed to enqueue regardless of
## the current queue size. This avoids wasting the time spent being accepted
## into merge windows, which would happen if the merge were to be bounced with
## a busy-reply that would subsequently be unwound through the entire merge chain.
disable_queue_limits_for_chained_merges bool default=true
## Whether the deadlock detector should be enabled or not. If disabled, it will
## still run, but it will never actually abort the process it is running in.
enable_dead_lock_detector bool default=false
## Whether to enable deadlock detector warnings in log or not. If enabled,
## warnings will be written even if dead lock detecting is not enabled.
enable_dead_lock_detector_warnings bool default=true
## Each thread registers how often it will at minimum register ticks (given that
## the system is not overloaded. If you are running Vespa on overloaded nodes,
## you can use this slack timeout to add to the thread timeouts in order to
## allow for more slack before dead lock detector kicks in. The value is in seconds.
dead_lock_detector_timeout_slack double default=240
## If set to 0, storage will attempt to auto-detect the number of VDS mount
## points to use. If set to a number, force this number. This number only makes
## sense on a storage node of course
disk_count int default=0 restart
## Configure persistence provider. Temporary here to test.
persistence_provider.type enum {STORAGE, DUMMY, RPC } default=STORAGE restart
persistence_provider.rpc.connectspec string default="tcp/localhost:27777" restart
## Whether or not to use the new metadata flow implementation. Default to not
## as it is currently in development and not even functional
switch_new_meta_data_flow bool default=false restart
## When the content layer receives a set of changed buckets from the persistence
## layer, it must recheck all of these. Each such recheck results in an
## operation scheduled against the persistence queust and since the total
## number of buckets to recheck may reach hundreds of thousands in a large
## system, we send these in chunks to avoid saturating the queues with
## operations.
bucket_rechecking_chunk_size int default=100
## If greater than zero, simulates added latency caused by CPU processing during
## full bucket info requests. The latency is added per batch of operations processed.
## Only useful for testing!
simulated_bucket_request_latency_msec int default=0
## If set, content node processes will use a B-tree backed bucket database implementation
## instead of the legacy Judy-based implementation.
use_content_node_btree_bucket_db bool default=true restart
## If non-zero, the bucket DB will be striped into 2^bits sub-databases, each handling
## a disjoint subset of the node's buckets, in order to reduce locking contention.
## Max value is unspecified, but will be clamped internally.
content_node_bucket_db_stripe_bits int default=4 restart
|
