| layout | title | permalink | redirect_from | ||
|---|---|---|---|---|---|
post |
CONFIGURATION |
/docs/configuration |
|
AIS configuration consists of cluster-wide (global) defaults and node-specific values - the latter includes node's own hostnames (or IP addresses) and mountpaths (disks).
Optionally and in addition, there's also a node-specific "config override" - a set of values that were changed for this node from global (inherited) defaults.
It is important to note that configuring cluster for production requires careful consideration. For example, AIS supports 3 (three) logical networks and will, therefore, benefit, performance-wise, if provisioned with up to 3 isolated physical networks or VLANs. The logical networks are:
- user (aka public)
- intra-cluster control
- intra-cluster data
with the corresponding JSON names, respectively:
hostnamehostname_intra_controlhostname_intra_data
For AIS Kubernetes deployments we recommended Cilium CNI.
- To enable an optional AIStore authentication server, execute
$ AUTH_ENABLED=true make deploy. For information on AuthN server, please see AuthN documentation. - In addition to AIStore - the storage cluster, you can also deploy aisfs - to access AIS objects as files, and AIS CLI - to monitor, configure and manage AIS nodes and buckets.
- AIS CLI is an easy-to-use command-line management tool supporting a growing number of commands and options (one of the first ones you may want to try could be
ais show cluster- show the state and status of an AIS cluster). The CLI is documented in the readme; getting started with it boils down to runningmake cliand following the prompts. - For more testing commands and options, please refer to the testing README.
- For
aisnodecommand-line options, see: command-line options. - For helpful links and/or background on Go, AWS, GCP, and Deep Learning: helpful links.
- And again, run
make helpto find out how to build, run, and test AIStore and tools.
The first thing to keep in mind is that there are 3 (three) separate, and separately maintained, pieces:
- Cluster configuration that comprises global defaults
- Node (local) configuration
- Node's local overrides of global defaults
Specifically:
In the documentation and in the code we can say "global configuration" or, same, "cluster configuration". The point, though, is that global config is replicated, versioned, checksummed, compressed, and - most importantly - applies to the entire cluster, all current (and future) node members.
Typically, when we deploy a new AIS cluster, we use configuration template that contains all the defaults - see, for example, JSON template. Configuration sections in this template, and the knobs within those sections, must be self-explanatory, and the majority of those, except maybe just a few, have pre-assigned default values.
To show and/or change global config, simply type one of:
# 1. show cluster config
$ ais show cluster config
# 2. show cluster config in JSON format
$ ais show cluster config --json
# 3. show cluster-wide defaults for all variables prefixed with "time"
$ ais show config cluster time
# or, same:
$ ais show cluster config time
PROPERTY VALUE
timeout.cplane_operation 2s
timeout.max_keepalive 4s
timeout.max_host_busy 20s
timeout.startup_time 1m
timeout.send_file_time 5m
timeout.transport_idle_term 4s
# 4. for all nodes in the cluster set startup timeout to 2 minutes
$ ais config cluster timeout.startup_time=2m
config successfully updatedUnlike global configuration that is replicated across all nodes there is also a node-specific configuration comprising:
- local config and log directories
- network configuration, including node's hostname(s) or IP addresses
- node's mountpaths
Since AIS supports n-way mirroring and erasure coding, we typically recommend not using LVMs and hardware RAIDs.
# ais show config CCDpt8088
PROPERTY VALUE DEFAULT
auth.enabled false -
backend.conf map[aws:map[] gcp:map[]] -
checksum.enable_read_range false -
...
...
(Hint: use `--type` to select the node config's type to show: 'cluster', 'local', 'all'.)
...
...$ ais show config CCDpt8088 --json | tail -20
"lastupdate_time": "2021-03-20 18:00:20.393881867 -0700 PDT m=+2907.143584987",
"uuid": "ZzCknLkMi",
"config_version": "3",
"confdir": "/ais",
"log_dir": "/tmp/ais/log",
"host_net": {
"hostname": "",
"hostname_intra_control": "",
"hostname_intra_data": "",
"port": "51081",
"port_intra_control": "51082",
"port_intra_data": "51083"
},
"fspaths": {"/ais/mp1":{},"/ais/mp2":{},"/ais/mp3":{},"/ais/mp4":{}},
"test_fspaths": {
"root": "/tmp/ais",
"count": 0,
"instance": 0
}# ais show config koLAt8081 --type local
PROPERTY VALUE
confdir /ais
log_dir /tmp/ais/log
host_net.hostname
host_net.hostname_intra_control
host_net.hostname_intra_data
host_net.port 51081
host_net.port_intra_control 51082
host_net.port_intra_data 51083
fspaths.paths /ais/mp1,/ais/mp2,/ais/mp3,/ais/mp4
test_fspaths.root /tmp/ais
test_fspaths.count 0
test_fspaths.instance 0Finally, each clustered node can individually override inherited defaults. For example:
# ais show config CCDpt8088 timeout
PROPERTY VALUE DEFAULT
timeout.cplane_operation 2s -
timeout.max_host_busy 20s -
timeout.max_keepalive 4s -
timeout.send_file_time 5m -
timeout.startup_time 1m -
timeout.transport_idle_term 4s -
# ais cluster configure CCDpt8088 timeout.startup_time=2m
config for node "CCDpt8088" successfully updated
# ais show config CCDpt8088 timeout
PROPERTY VALUE DEFAULT
timeout.cplane_operation 2s -
timeout.max_host_busy 20s -
timeout.max_keepalive 4s -
timeout.send_file_time 5m -
timeout.startup_time 2m 1m
timeout.transport_idle_term 4s -Notice the DEFAULT column above where - indicates that the corresponding value is inherited and remains unchanged.
Rest of this document is structured as follows:
- Basics
- Startup override
- Managing mountpaths
- Disabling extended attributes
- Enabling HTTPS
- Filesystem Health Checker
- Networking
- Reverse proxy
- Curl examples
- CLI examples
AIS production deployment, in particular, requires careful consideration of at least some of the configurable aspects. For example, AIS supports 3 (three) logical networks and will, therefore, benefit, performance-wise, if provisioned with up to 3 isolated physical networks or VLANs. The logical networks are: user (aka public), intra-cluster control, and intra-cluster data - the corresponding JSON names are, respectively: hostname, hostname_intra_control, and hostname_intra_data.
The following picture illustrates one section of the configuration template that, in part, includes listening port:
Further, test_fspaths section (see below) corresponds to a single local filesystem being partitioned between both local and Cloud buckets. In other words, the test_fspaths configuration option is intended strictly for development.
In production, we use an alternative configuration called fspaths: the section of the config that includes a number of local directories, whereby each directory is based on a different local filesystem.
For fspath and mountpath terminology and details, please see section Managing Mountpaths in this document.
An example of 12 fspaths (and 12 local filesystems) follows below:
First, some basic facts:
- AIS cluster is a collection of nodes - members of the cluster.
- A node can be an AIS proxy (aka gateway) or an AIS target.
- In either case, HTTP request to read (get) or write (set) specific node's configuration will have
/v1/daemonin its URL path. - The capability to carry out cluster-wide configuration updates is also supported. The corresponding HTTP URL will have
/v1/clusterin its path.
Both
daemonandclusterare the two RESTful resource abstractions supported by the API. Please see AIS API for naming conventions, RESTful resources, as well as API reference and details.
- To get the node's up-to-date configuration, execute:
$ ais show config <daemon-ID>This will display all configuration sections and all the named knobs - i.e., configuration variables and their current values.
Most configuration options can be updated either on an individual (target or proxy) daemon, or the entire cluster. Some configurations are "overridable" and can be configured on a per-daemon basis. Some of these are shown in the table below.
For examples and alternative ways to format configuration-updating requests, please see the examples below.
Following is a table-summary that contains a subset of all settable knobs:
| Option name | Overridable | Default value | Description |
|---|---|---|---|
ec.data_slices |
No | 2 |
Represents the number of fragments an object is broken into (in the range [2, 100]) |
ec.disk_only |
No | false |
If true, EC uses local drives for all operations. If false, EC automatically chooses between memory and local drives depending on the current memory load |
ec.enabled |
No | false |
Enables or disables data protection |
ec.objsize_limit |
No | 262144 |
Indicated the minimum size of an object in bytes that is erasure encoded. Smaller objects are replicated |
ec.parity_slices |
No | 2 |
Represents the number of redundant fragments to provide protection from failures (in the range [2, 32]) |
ec.compression |
No | "never" |
LZ4 compression parameters used when EC sends its fragments and replicas over network. Values: "never" - disables, "always" - compress all data, or a set of rules for LZ4, e.g "ratio=1.2" means enable compression from the start but disable when average compression ratio drops below 1.2 to save CPU resources |
mirror.burst_buffer |
No | 512 |
the maximum length of the queue of objects to be mirrored. When the queue length exceeds the value, a target may skip creating replicas for new objects |
mirror.copies |
No | 1 |
the number of local copies of an object |
mirror.enabled |
No | false |
If true, for every object PUT a target creates object replica on another mountpath. Later, on object GET request, loadbalancer chooses a mountpath with lowest disk utilization and reads the object from it |
mirror.util_thresh |
No | 20 |
If mirroring is enabled, loadbalancer chooses an object replica to read but only if main object's mountpath utilization exceeds the replica' s mountpath utilization by this value. Main object's mountpath is the mountpath used to store the object when mirroring is disabled |
rebalance.dest_retry_time |
No | 2m |
If a target does not respond within this interval while rebalance is running the target is excluded from rebalance process |
rebalance.enabled |
No | true |
Enables and disables automatic rebalance after a target receives the updated cluster map. If the (automated rebalancing) option is disabled, you can still use the REST API (PUT {"action": "start", "value": {"kind": "rebalance"}} v1/cluster) to initiate cluster-wide rebalancing |
rebalance.multiplier |
No | 4 |
A tunable that can be adjusted to optimize cluster rebalancing time (advanced usage only) |
rebalance.quiescent |
No | 20s |
Rebalance moves to the next stage or starts the next batch of objects when no objects are received during this time interval |
versioning.enabled |
No | true |
Enables and disables versioning. For the supported 3rd party backends, versioning is on only when it enabled for (and supported by) the specific backend |
versioning.validate_warm_get |
No | false |
If false, a target returns a requested object immediately if it is cached. If true, a target fetches object's version(via HEAD request) from Cloud and if the received version mismatches locally cached one, the target redownloads the object and then returns it to a client |
checksum.enable_read_range |
Yes | false |
See Supported Checksums and Brief Theory of Operations |
checksum.type |
Yes | xxhash |
Checksum type. Please see Supported Checksums and Brief Theory of Operations |
checksum.validate_cold_get |
Yes | true |
Please see Supported Checksums and Brief Theory of Operations |
checksum.validate_warm_get |
Yes | false |
See Supported Checksums and Brief Theory of Operations |
client.client_long_timeout |
Yes | 30m |
Default long client timeout |
client.client_timeout |
Yes | 10s |
Default client timeout |
client.list_timeout |
Yes | 2m |
Client list objects timeout |
compression.block_size |
Yes | 262144 |
Maximum data block size used by LZ4, greater values may increase compression ration but requires more memory. Value is one of 64KB, 256KB(AIS default), 1MB, and 4MB |
disk.disk_util_high_wm |
Yes | 80 |
Operations that implement self-throttling mechanism, e.g. LRU, turn on the maximum throttle if disk utilization is higher than disk_util_high_wm |
disk.disk_util_low_wm |
Yes | 60 |
Operations that implement self-throttling mechanism, e.g. LRU, do not throttle themselves if disk utilization is below disk_util_low_wm |
disk.iostat_time_long |
Yes | 2s |
The interval that disk utilization is checked when disk utilization is below disk_util_low_wm. |
disk.iostat_time_short |
Yes | 100ms |
Used instead of iostat_time_long when disk utilization reaches disk_util_high_wm. If disk utilization is between disk_util_high_wm and disk_util_low_wm, a proportional value between iostat_time_short and iostat_time_long is used. |
distributed_sort.call_timeout |
Yes | "10m" |
a maximum time a target waits for another target to respond |
distributed_sort.compression |
Yes | "never" |
LZ4 compression parameters used when dSort sends its shards over network. Values: "never" - disables, "always" - compress all data, or a set of rules for LZ4, e.g "ratio=1.2" means enable compression from the start but disable when average compression ratio drops below 1.2 to save CPU resources |
distributed_sort.default_max_mem_usage |
Yes | "80%" |
a maximum amount of memory used by running dSort. Can be set as a percent of total memory(e.g 80%) or as the number of bytes(e.g, 12G) |
distributed_sort.dsorter_mem_threshold |
Yes | "100GB" |
minimum free memory threshold which will activate specialized dsorter type which uses memory in creation phase - benchmarks shows that this type of dsorter behaves better than general type |
distributed_sort.duplicated_records |
Yes | "ignore" |
what to do when duplicated records are found: "ignore" - ignore and continue, "warn" - notify a user and continue, "abort" - abort dSort operation |
distributed_sort.ekm_malformed_line |
Yes | "abort" |
what to do when extraction key map notices a malformed line: "ignore" - ignore and continue, "warn" - notify a user and continue, "abort" - abort dSort operation |
distributed_sort.ekm_missing_key |
Yes | "abort" |
what to do when extraction key map have a missing key: "ignore" - ignore and continue, "warn" - notify a user and continue, "abort" - abort dSort operation |
distributed_sort.missing_shards |
Yes | "ignore" |
what to do when missing shards are detected: "ignore" - ignore and continue, "warn" - notify a user and continue, "abort" - abort dSort operation |
fshc.enabled |
Yes | true |
Enables and disables filesystem health checker (FSHC) |
log.level |
Yes | 3 |
Set global logging level. The greater number the more verbose log output |
lru.capacity_upd_time |
Yes | 10m |
Determines how often AIStore updates filesystem usage |
lru.dont_evict_time |
Yes | 120m |
LRU does not evict an object which was accessed less than dont_evict_time ago |
lru.enabled |
Yes | true |
Enables and disabled the LRU |
lru.highwm |
Yes | 90 |
LRU starts immediately if a filesystem usage exceeds the value |
lru.lowwm |
Yes | 75 |
If filesystem usage exceeds highwm LRU tries to evict objects so the filesystem usage drops to lowwm |
periodic.notif_time |
Yes | 30s |
An interval of time to notify subscribers (IC members) of the status and statistics of a given asynchronous operation (such as Download, Copy Bucket, etc.) |
periodic.stats_time |
Yes | 10s |
A housekeeping time interval to periodically update and log internal statistics, remove/rotate old logs, check available space (and run LRU xaction if need be), etc. |
resilver.enabled |
Yes | true |
Enables and disables automatic reresilver after a mountpath has been added or removed. If the (automated resilvering) option is disabled, you can still use the REST API (PUT {"action": "start", "value": {"kind": "resilver", "node": targetID}} v1/cluster) to initiate resilvering |
timeout.max_host_busy |
Yes | 20s |
Maximum latency of control-plane operations that may involve receiving new bucket metadata and associated processing |
timeout.send_file_time |
Yes | 5m |
Timeout for sending/receiving an object from another target in the same cluster |
timeout.transport_idle_term |
Yes | 4s |
Max idle time to temporarily teardown long-lived intra-cluster connection |
AIS command-line allows to override configuration at AIS node's startup. For example:
$ aisnode -config=/etc/ais.json -local_config=/etc/ais_local.json -role=target -config_custom="client.timeout=13s,log.level=4"As shown above, the CLI option in-question is: confjson.
Its value is a JSON-formatted map of string names and string values.
By default, the config provided in config_custom will be persisted on the disk.
To make it transient either add -transient=true flag or add additional JSON entry:
$ aisnode -config=/etc/ais.json -local_config=/etc/ais_local.json -role=target -transient=true -config_custom="client.timeout=13s, transient=true"Another example. To override locally-configured address of the primary proxy, run:
$ aisnode -config=/etc/ais.json -local_config=/etc/ais_local.json -role=target -config_custom="proxy.primary_url=http://G"
# where G denotes the designated primary's hostname and port.To achieve the same on temporary basis, add -transient=true as follows:
$ aisnode -config=/etc/ais.json -local_config=/etc/ais_local.json -role=target -config_custom="proxy.primary_url=http://G"Please see AIS command-line for other command-line options and details.
- Mountpath - is a single disk or a volume (a RAID) formatted with a local filesystem of choice, and a local directory that AIS can fully own and utilize (to store user data and system metadata). Note that any given disk (or RAID) can have (at most) one mountpath (meaning no disk sharing) and mountpath directories cannot be nested. Further:
- a mountpath can be temporarily disabled and (re)enabled;
- a mountpath can also be detached and (re)attached, thus effectively supporting growth and "shrinkage" of local capacity;
- it is safe to execute the 4 listed operations (enable, disable, attach, detach) at any point during runtime;
- in a typical deployment, the total number of mountpaths would compute as a direct product of (number of storage targets) x (number of disks in each target).
Configuration option fspaths specifies the list of local mountpath directories. Each configured fspath is, simply, a local directory that provides the basis for AIS mountpath.
In regards non-sharing of disks between mountpaths: for development we make an exception, such that multiple mountpaths are actually allowed to share a disk and coexist within a single filesystem. This is done strictly for development convenience, though.
AIStore HTTP API makes it possible to list, add, remove, enable, and disable a fspath (and, therefore, the corresponding local filesystem) at runtime. Filesystem's health checker (FSHC) monitors the health of all local filesystems: a filesystem that "accumulates" I/O errors will be disabled and taken out, as far as the AIStore built-in mechanism of object distribution. For further details about FSHC, please refer to FSHC readme.
To make sure that AIStore does not utilize xattrs, configure checksum.type=none, versioning.enabled=true,
and md_write=never for all targets in a AIStore cluster.
This can be done via the common configuration "part" that'd be further used to deploy the cluster.
Extended attributes can be disabled on per bucket basis. To do this, turn off saving metadata to disks (CLI):
$ ais bucket props ais://mybucket md_write=never
Bucket props successfully updated
"md_write" set to: "never" (was: "")Disable extended attributes only if you need fast and temporary storage. Without xattrs, a node loses its objects after the node reboots. If extended attributes are disabled globally when deploying a cluster, node IDs are not permanent and a node can change its ID after it restarts.
To switch from HTTP protocol to an encrypted HTTPS, configure net.http.use_https=true and modify net.http.server_crt and net.http.server_key values so they point to your OpenSSL certificate and key files respectively (see AIStore configuration).
Default installation enables filesystem health checker component called FSHC. FSHC can be also disabled via section "fshc" of the configuration.
When enabled, FSHC gets notified on every I/O error upon which it performs extensive checks on the corresponding local filesystem. One possible outcome of this health-checking process is that FSHC disables the faulty filesystems leaving the target with one filesystem less to distribute incoming data.
Please see FSHC readme for further details.
In addition to user-accessible public network, AIStore will optionally make use of the two other networks: internal (or intra-cluster) and replication. If configured via the net section of the configuration, the intra-cluster network is utilized for latency-sensitive control plane communications including keep-alive and metasync. The replication network is used, as the name implies, for a variety of replication workloads.
All the 3 (three) networking options are enumerated here.
AIStore gateway can act as a reverse proxy vis-à-vis AIStore storage targets. This functionality is limited to GET requests only and must be used with caution and consideration. Related configuration variable is called rproxy - see sub-section http of the section net. For further details, please refer to this readme.
The following assumes that G and T are the (hostname:port) of one of the deployed gateways (in a given AIS cluster) and one of the targets, respectively.
- Set the stats logging interval to 1 second
$ curl -i -X PUT -H 'Content-Type: application/json' -d '{"action": "set-config","name": "periodic.stats_time", "value": "1s"}' 'http://G/v1/cluster'or, same:
$ curl -i -X PUT 'http://G/v1/cluster/set-config?periodic.stats_time=1s'Notice the two alternative ways to form the requests.
- Set the stats logging interval to 2 minutes
$ curl -i -X PUT -H 'Content-Type: application/json' -d '{"action": "set-config","name": "periodic.stats_time", "value": "2m"}' 'http://G/v1/cluster'- Set the default number of n-way copies to 4 (can still be redefined on a per-bucket basis)
$ curl -i -X PUT -H 'Content-Type: application/json' -d '{"action": "set-config","name": "mirror.copies", "value": "4"}' 'http://G/v1/cluster'
# or, same using CLI:
$ ais config cluster mirror.copies 4- Set log level = 1
$ curl -i -X PUT -H 'Content-Type: application/json' -d '{"action": "set-config","name": "log.level", "value": "1"}' 'http://T/v1/daemon'
# or, same:
$ curl -i -X PUT 'http://T/v1/daemon/set-config?log.level=1'
# or, same using CLI (assuming the node in question is t[tZktGpbM]):
$ ais config node t[tZktGpbM] log.level 1AIS CLI is an integrated management-and-monitoring command line tool. The following CLI command sequence, first - finds out all AIS knobs that contain substring "time" in their names, second - modifies list_timeout from 2 minutes to 5 minutes, and finally, displays the modified value:
$ ais show config --type all 844974_8080 --json | jq '.timeout.list_timeout'
"2m"
$ ais config cluster timeout.list_timeout=5m
Config has been updated successfully.
$ ais show config --type all 844974_8080 --json | jq '.timeout.list_timeout'
"5m"The example above demonstrates cluster-wide configuration update but note: single-node updates are also supported.
- Set
periodic.stats_time= 1 minute,periodic.iostat_time_long= 4 seconds
$ ais config cluster periodic.stats_time=1m disk.iostat_time_long=4sAIS configuration includes a section called disk. The disk in turn contains several knobs - one of those knobs is disk.iostat_time_long, another - disk.disk_util_low_wm. To update one or both of those named variables on all or one of the clustered nodes, you could:
- Set
disk.iostat_time_long= 3 seconds,disk.disk_util_low_wm= 40 percent on daemon with IDtarget1
$ ais config node target1 periodic.stats_time=1m disk.iostat_time_long=4s