The thanos receive command implements the Prometheus Remote Write API. It builds on top of existing Prometheus TSDB and retains its usefulness while extending its functionality with long-term-storage, horizontal scalability, and downsampling. Prometheus instances are configured to continuously write metrics to it, and then Thanos Receive uploads TSDB blocks to an object storage bucket every 2 hours by default. Thanos Receive exposes the StoreAPI so that Thanos Queriers can query received metrics in real-time.
We recommend this component to users who can only push into a Thanos due to air-gapped, or egress only environments. Please note the various pros and cons of pushing metrics.
Thanos Receive supports multi-tenancy by using labels. See Multi-tenancy documentation here.
Thanos Receive supports ingesting exemplars via remote-write. By default, the exemplars are silently discarded as --tsdb.max-exemplars is set to 0. To enable exemplars storage, set the --tsdb.max-exemplars flag to a non-zero value. It exposes the ExemplarsAPI so that the Thanos Queriers can query the stored exemplars. Take a look at the documentation for exemplars storage in Prometheus to know more about it.
For more information please check out initial design proposal. For further information on tuning Prometheus Remote Write see remote write tuning document.
NOTE: As the block producer it’s important to set correct “external labels” that will identify data block across Thanos clusters. See external labels docs for details.
The Receive component currently supports two algorithms for distributing timeseries across Receive nodes and can be set using the receive.hashrings-algorithm flag.
The Ketama algorithm is a consistent hashing scheme which enables stable scaling of Receivers without the drawbacks of the hashmod algorithm. This is the recommended algorithm for all new installations.
If you are using the hashmod algorithm and wish to migrate to ketama, the simplest and safest way would be to set up a new pool receivers with ketama hashrings and start remote-writing to them. Provided you are on the latest Thanos version, old receivers will flush their TSDBs after the configured retention period and will upload blocks to object storage. Once you have verified that is done, decommission the old receivers.
This algorithm uses a hashmod function over all labels to decide which receiver is responsible for a given timeseries. This is the default algorithm due to historical reasons. However, its usage for new Receive installations is discouraged since adding new Receiver nodes leads to series churn and memory usage spikes.
By default, Receivers replicate data using Protobuf over gRPC. Deserializing protobuf-encoded messages can be resource-intensive and cause significant GC pressure. Alternatively, you can use Cap’N Proto for replication encoding and as the RPC framework.
In order to enable this mode, you can use the receive.replication-protocol=capnproto option on the receiver. Thanos will try to infer the Cap’N Proto address of each peer in the hashring using the existing gRPC address. You can also explicitly set the Cap’N Proto as follows:
[
{
"endpoints": [
{"address": "node-1:10901", "capnproto_address": "node-1:19391"},
{"address": "node-2:10901", "capnproto_address": "node-2:19391"},
{"address": "node-3:10901", "capnproto_address": "node-3:19391"}
]
}
]
The Thanos Receive Controller project aims to automate hashring management when running Thanos in Kubernetes. In combination with the Ketama hashring algorithm, this controller can also be used to keep hashrings up to date when Receivers are scaled automatically using an HPA or Keda.
Thanos Receive supports getting TSDB stats using the /api/v1/status/tsdb endpoint. Use the THANOS-TENANT HTTP header to get stats for individual Tenants. Use the limit query parameter to tweak the number of stats to return (the default is 10). The output format of the endpoint is compatible with Prometheus API.
Note that each Thanos Receive will only expose local stats and replicated series will not be included in the response.
Tenants in Receivers are created dynamically and do not need to be provisioned upfront. When a new value is detected in the tenant HTTP header, Receivers will provision and start managing an independent TSDB for that tenant. TSDB blocks that are sent to S3 will contain a unique tenant_id label which can be used to compact blocks independently for each tenant.
A Receiver will automatically decommission a tenant once new samples have not been seen for longer than the --tsdb.retention period configured for the Receiver. The tenant decommission process includes flushing all in-memory samples for that tenant to disk, sending all unsent blocks to S3, and removing the tenant TSDB from the filesystem. If a tenant receives new samples after being decommissioned, a new TSDB will be created for the tenant.
Note that because of the built-in decommissioning process, the semantic of the --tsdb.retention flag in the Receiver is different than the one in Prometheus. For Receivers, --tsdb.retention=t indicates that the data for a tenant will be kept for t amount of time, whereas in Prometheus, --tsdb.retention=t denotes that the last t duration of data will be maintained in TSDB. In other words, Prometheus will keep the last t duration of data even when it stops getting new samples.
thanos receive \
--tsdb.path "/path/to/receive/data/dir" \
--grpc-address 0.0.0.0:10907 \
--http-address 0.0.0.0:10909 \
--receive.replication-factor 1 \
--label "receive_replica=\"0\"" \
--label "receive_cluster=\"eu1\"" \
--receive.local-endpoint 127.0.0.1:10907 \
--receive.hashrings-file ./data/hashring.json \
--remote-write.address 0.0.0.0:10908 \
--objstore.config-file "bucket.yml"
The example of remote_write Prometheus configuration:
remote_write:
- url: http://<thanos-receive-container-ip>:10908/api/v1/receive
where <thanos-receive-containter-ip> is an IP address reachable by Prometheus Server.
The example content of bucket.yml:
type: GCS
config:
bucket: ""
service_account: ""
use_grpc: false
grpc_conn_pool_size: 0
http_config:
idle_conn_timeout: 0s
response_header_timeout: 0s
insecure_skip_verify: false
tls_handshake_timeout: 0s
expect_continue_timeout: 0s
max_idle_conns: 0
max_idle_conns_per_host: 0
max_conns_per_host: 0
tls_config:
ca_file: ""
cert_file: ""
key_file: ""
server_name: ""
insecure_skip_verify: false
disable_compression: false
chunk_size_bytes: 0
max_retries: 0
prefix: ""
The example content of hashring.json:
[
{
"endpoints": [
"127.0.0.1:10907",
"127.0.0.1:11907",
"127.0.0.1:12907"
]
}
]
With such configuration any receive listens for remote write on <ip>10908/api/v1/receive and will forward to correct one in hashring if needed for tenancy and replication.
It is possible to only match certain tenants inside of a hashring file. For example:
[
{
"tenants": ["foobar"],
"endpoints": [
"127.0.0.1:1234",
"127.0.0.1:12345",
"127.0.0.1:1235"
]
}
]
The specified endpoints will be used if the tenant is set to foobar. It is possible to use glob matching through the parameter tenant_matcher_type. It can have the value glob. In this case, the strings inside the array are taken as glob patterns and matched against the tenant inside of a remote-write request. For instance:
[
{
"tenants": ["foo*"],
"tenant_matcher_type": "glob",
"endpoints": [
"127.0.0.1:1234",
"127.0.0.1:12345",
"127.0.0.1:1235"
]
}
]
This will still match the tenant foobar and any other tenant which begins with the letters foo.
In order to ensure even spread for replication over nodes in different availability-zones, you can choose to include az definition in your hashring config. If we for example have a 6 node cluster, spread over 3 different availability zones; A, B and C, we could use the following example hashring.json:
[
{
"endpoints": [
{
"address": "127.0.0.1:10907",
"az": "A"
},
{
"address": "127.0.0.1:11907",
"az": "B"
},
{
"address": "127.0.0.1:12907",
"az": "C"
},
{
"address": "127.0.0.1:13907",
"az": "A"
},
{
"address": "127.0.0.1:14907",
"az": "B"
},
{
"address": "127.0.0.1:15907",
"az": "C"
}
]
}
]
This is only supported for the Ketama algorithm.
NOTE: This feature is made available from v0.32 onwards. Receive can still operate with endpoints set to an array of IP strings in ketama mode. But to use AZ-aware hashring, you would need to migrate your existing hashring (and surrounding automation) to the new JSON structure mentioned above.
Thanos Receive has some limits and gates that can be configured to control resource usage. Here’s the difference between limits and gates:
To configure the gates and limits you can use one of the two options:
--receive.limits-config-file=<file-path>: where <file-path> is the path to the YAML file. Any modification to the indicated file will trigger a configuration reload. If the updated configuration is invalid an error will be logged and it won’t replace the previous valid configuration.--receive.limits-config=<content>: where <content> is the content of YAML file.By default all the limits and gates are disabled. These options should be added to the routing-receivers when using the Routing Receive and Ingesting Receive.
The configuration file follows a few standards:
0 (zero) is used to explicitly define “there is no limit” (infinite limit).default section) or global limits (in the global section), a value that is not present means “no limit”.tenants section), a value that is not present means they are the same as the default.All the configuration for the remote write endpoint of Receive is contained in the write key. Inside it there are 3 subsections:
global: limits, gates and/or options that are applied considering all the requests.default: the default values for limits in case a given tenant doesn’t have any specified.tenants: the limits for a given tenant.For a Receive instance with configuration like below, it’s understood that:
meta_monitoring_.* options in global.acme tenant and partially for the ajax tenant).acme has no request limits, but has a higher head_series limit.ajax has a request series limit of 50000 and samples limit of 500. Their request size bytes limit is inherited from the default, 1024 bytes. Their head series are also inherited from default i.e, 1000.The next sections explain what each configuration value means.
write:
global:
max_concurrency: 30
meta_monitoring_url: "http://localhost:9090"
meta_monitoring_limit_query: "sum(prometheus_tsdb_head_series) by (tenant)"
default:
request:
size_bytes_limit: 1024
series_limit: 1000
samples_limit: 10
head_series_limit: 1000
tenants:
acme:
request:
size_bytes_limit: 0
series_limit: 0
samples_limit: 0
head_series_limit: 2000
ajax:
request:
series_limit: 50000
samples_limit: 500
IMPORTANT: this feature is experimental and a work-in-progress. It might change in the near future, i.e. configuration might move to a file (to allow easy configuration of different request limits per tenant) or its structure could change.
Thanos Receive supports setting limits on the incoming remote write request sizes. These limits should help you to prevent a single tenant from being able to send big requests and possibly crash the Receive.
These limits are applied per request and can be configured within the request key:
size_bytes_limit: the maximum body size.series_limit: the maximum amount of series in a single remote write request.samples_limit: the maximum amount of samples in a single remote write request (summed from all series).Any request above these limits will cause an 413 HTTP response (Entity Too Large) and should not be retried without modifications.
Currently a 413 HTTP response will cause data loss at the client, as none of them (Prometheus included) will break down 413 responses into smaller requests. The recommendation is to monitor these errors in the client and contact the owners of your Receive instance for more information on its configured limits.
Future work that can improve this scenario:
By default, all these limits are disabled.
The available request gates in Thanos Receive can be configured within the global key:
max_concurrency: the maximum amount of remote write requests that will be concurrently worked on. Any request request that would exceed this limit will be accepted, but wait until the gate allows it to be processed.Thanos Receive, in Router or RouterIngestor mode, supports limiting tenant active (head) series to maintain the system’s stability. It uses any Prometheus Query API compatible meta-monitoring solution that consumes the metrics exposed by all receivers in the Thanos system. Such query endpoint allows getting the scrape time seconds old number of all active series per tenant, which is then compared with a configured limit before ingesting any tenant’s remote write request. In case a tenant has gone above the limit, their remote write requests fail fully.
Every Receive Router/RouterIngestor node, queries meta-monitoring for active series of all tenants, every 15 seconds, and caches the results in a map. This cached result is used to limit all incoming remote write requests.
To use the feature, one should specify the following limiting config options:
Under global:
meta_monitoring_url: Specifies Prometheus Query API compatible meta-monitoring endpoint.meta_monitoring_limit_query: Option to specify PromQL query to execute against meta-monitoring. If not specified it is set to sum(prometheus_tsdb_head_series) by (tenant) by default.meta_monitoring_http_client: Optional YAML field specifying HTTP client config for meta-monitoring.Under default and per tenant:
head_series_limit: Specifies the total number of active (head) series for any tenant, across all replicas (including data replication), allowed by Thanos Receive. Set to 0 for unlimited.NOTE:
Instead of spawning a new goroutine each time the Receiver forwards a request to another node, it spawns a fixed number of goroutines (workers) that perform the work. This allows avoiding spawning potentially tens or even hundred thousand goroutines if someone starts sending a lot of small requests.
This number of workers is controlled by --receive.forward.async-workers=.
Please see the metric thanos_receive_forward_delay_seconds to see if you need to increase the number of forwarding workers.
The following formula is used for calculating quorum:
// writeQuorum returns minimum number of replicas that has to confirm write success before claiming replication success.
func (h *Handler) writeQuorum() int {
// NOTE(GiedriusS): this is here because otherwise RF=2 doesn't make sense as all writes
// would need to succeed all the time. Another way to think about it is when migrating
// from a Sidecar based setup with 2 Prometheus nodes to a Receiver setup, we want to
// keep the same guarantees.
if h.options.ReplicationFactor == 2 {
return 1
}
return int((h.options.ReplicationFactor / 2) + 1)
}
So, if the replication factor is 2 then at least one write must succeed. With RF=3, two writes must succeed, and so on.
usage: thanos receive [<flags>]
Accept Prometheus remote write API requests and write to local tsdb.
Flags:
--auto-gomemlimit.ratio=0.9
The ratio of reserved GOMEMLIMIT memory to the
detected maximum container or system memory.
--enable-auto-gomemlimit Enable go runtime to automatically limit memory
consumption.
--enable-feature= ... Comma separated experimental feature names
to enable. The current list of features is
metric-names-filter.
--grpc-address="0.0.0.0:10901"
Listen ip:port address for gRPC endpoints
(StoreAPI). Make sure this address is routable
from other components.
--grpc-grace-period=2m Time to wait after an interrupt received for
GRPC Server.
--grpc-server-max-connection-age=60m
The grpc server max connection age. This
controls how often to re-establish connections
and redo TLS handshakes.
--grpc-server-tls-cert="" TLS Certificate for gRPC server, leave blank to
disable TLS
--grpc-server-tls-client-ca=""
TLS CA to verify clients against. If no
client CA is specified, there is no client
verification on server side. (tls.NoClientCert)
--grpc-server-tls-key="" TLS Key for the gRPC server, leave blank to
disable TLS
--grpc-server-tls-min-version="1.3"
TLS supported minimum version for gRPC server.
If no version is specified, it'll default to
1.3. Allowed values: ["1.0", "1.1", "1.2",
"1.3"]
--hash-func= Specify which hash function to use when
calculating the hashes of produced files.
If no function has been specified, it does not
happen. This permits avoiding downloading some
files twice albeit at some performance cost.
Possible values are: "", "SHA256".
-h, --help Show context-sensitive help (also try
--help-long and --help-man).
--http-address="0.0.0.0:10902"
Listen host:port for HTTP endpoints.
--http-grace-period=2m Time to wait after an interrupt received for
HTTP Server.
--http.config="" [EXPERIMENTAL] Path to the configuration file
that can enable TLS or authentication for all
HTTP endpoints.
--label=key="value" ... External labels to announce. This flag will be
removed in the future when handling multiple
tsdb instances is added.
--log.format=logfmt Log format to use. Possible options: logfmt or
json.
--log.level=info Log filtering level.
--objstore.config=<content>
Alternative to 'objstore.config-file'
flag (mutually exclusive). Content of
YAML file that contains object store
configuration. See format details:
https://thanos.io/tip/thanos/storage.md/#configuration
--objstore.config-file=<file-path>
Path to YAML file that contains object
store configuration. See format details:
https://thanos.io/tip/thanos/storage.md/#configuration
--receive.capnproto-address="0.0.0.0:19391"
Address for the Cap'n Proto server.
--receive.default-tenant-id="default-tenant"
Default tenant ID to use when none is provided
via a header.
--receive.forward.async-workers=5
Number of concurrent workers processing
forwarding of remote-write requests.
--receive.grpc-compression=snappy
Compression algorithm to use for gRPC requests
to other receivers. Must be one of: snappy,
none
--receive.hashrings=<content>
Alternative to 'receive.hashrings-file' flag
(lower priority). Content of file that contains
the hashring configuration.
--receive.hashrings-algorithm=hashmod
The algorithm used when distributing series in
the hashrings. Must be one of hashmod, ketama.
Will be overwritten by the tenant-specific
algorithm in the hashring config.
--receive.hashrings-file=<path>
Path to file that contains the hashring
configuration. A watcher is initialized
to watch changes and update the hashring
dynamically.
--receive.hashrings-file-refresh-interval=5m
Refresh interval to re-read the hashring
configuration file. (used as a fallback)
--receive.local-endpoint=RECEIVE.LOCAL-ENDPOINT
Endpoint of local receive node. Used to
identify the local node in the hashring
configuration. If it's empty AND hashring
configuration was provided, it means that
receive will run in RoutingOnly mode.
--receive.relabel-config=<content>
Alternative to 'receive.relabel-config-file'
flag (mutually exclusive). Content of YAML file
that contains relabeling configuration.
--receive.relabel-config-file=<file-path>
Path to YAML file that contains relabeling
configuration.
--receive.replica-header="THANOS-REPLICA"
HTTP header specifying the replica number of a
write request.
--receive.replication-factor=1
How many times to replicate incoming write
requests.
--receive.replication-protocol=protobuf
The protocol to use for replicating
remote-write requests. One of protobuf,
capnproto
--receive.split-tenant-label-name=""
Label name through which the request will
be split into multiple tenants. This takes
precedence over the HTTP header.
--receive.tenant-certificate-field=
Use TLS client's certificate field to
determine tenant for write requests.
Must be one of organization, organizationalUnit
or commonName. This setting will cause the
receive.tenant-header flag value to be ignored.
--receive.tenant-header="THANOS-TENANT"
HTTP header to determine tenant for write
requests.
--receive.tenant-label-name="tenant_id"
Label name through which the tenant will be
announced.
--remote-write.address="0.0.0.0:19291"
Address to listen on for remote write requests.
--remote-write.client-server-name=""
Server name to verify the hostname
on the returned TLS certificates. See
https://tools.ietf.org/html/rfc4366#section-3.1
--remote-write.client-tls-ca=""
TLS CA Certificates to use to verify servers.
--remote-write.client-tls-cert=""
TLS Certificates to use to identify this client
to the server.
--remote-write.client-tls-key=""
TLS Key for the client's certificate.
--remote-write.client-tls-secure
Use TLS when talking to the other receivers.
--remote-write.client-tls-skip-verify
Disable TLS certificate verification when
talking to the other receivers i.e self signed,
signed by fake CA.
--remote-write.server-tls-cert=""
TLS Certificate for HTTP server, leave blank to
disable TLS.
--remote-write.server-tls-client-ca=""
TLS CA to verify clients against. If no
client CA is specified, there is no client
verification on server side. (tls.NoClientCert)
--remote-write.server-tls-key=""
TLS Key for the HTTP server, leave blank to
disable TLS.
--remote-write.server-tls-min-version="1.3"
TLS version for the gRPC server, leave blank
to default to TLS 1.3, allow values: ["1.0",
"1.1", "1.2", "1.3"]
--request.logging-config=<content>
Alternative to 'request.logging-config-file'
flag (mutually exclusive). Content
of YAML file with request logging
configuration. See format details:
https://thanos.io/tip/thanos/logging.md/#configuration
--request.logging-config-file=<file-path>
Path to YAML file with request logging
configuration. See format details:
https://thanos.io/tip/thanos/logging.md/#configuration
--store.limits.request-samples=0
The maximum samples allowed for a single
Series request, The Series call fails if
this limit is exceeded. 0 means no limit.
NOTE: For efficiency the limit is internally
implemented as 'chunks limit' considering each
chunk contains a maximum of 120 samples.
--store.limits.request-series=0
The maximum series allowed for a single Series
request. The Series call fails if this limit is
exceeded. 0 means no limit.
--tracing.config=<content>
Alternative to 'tracing.config-file' flag
(mutually exclusive). Content of YAML file
with tracing configuration. See format details:
https://thanos.io/tip/thanos/tracing.md/#configuration
--tracing.config-file=<file-path>
Path to YAML file with tracing
configuration. See format details:
https://thanos.io/tip/thanos/tracing.md/#configuration
--tsdb.allow-overlapping-blocks
Allow overlapping blocks, which in turn enables
vertical compaction and vertical query merge.
Does not do anything, enabled all the time.
--tsdb.max-exemplars=0 Enables support for ingesting exemplars and
sets the maximum number of exemplars that will
be stored per tenant. In case the exemplar
storage becomes full (number of stored
exemplars becomes equal to max-exemplars),
ingesting a new exemplar will evict the oldest
exemplar from storage. 0 (or less) value of
this flag disables exemplars storage.
--tsdb.max-retention-bytes=0
Maximum number of bytes that can be stored for
blocks. A unit is required, supported units: B,
KB, MB, GB, TB, PB, EB. Ex: "512MB". Based on
powers-of-2, so 1KB is 1024B.
--tsdb.no-lockfile Do not create lockfile in TSDB data directory.
In any case, the lockfiles will be deleted on
next startup.
--tsdb.path="./data" Data directory of TSDB.
--tsdb.retention=15d How long to retain raw samples on local
storage. 0d - disables the retention
policy (i.e. infinite retention).
For more details on how retention is
enforced for individual tenants, please
refer to the Tenant lifecycle management
section in the Receive documentation:
https://thanos.io/tip/components/receive.md/#tenant-lifecycle-management
--tsdb.too-far-in-future.time-window=0s
Configures the allowed time window for
ingesting samples too far in the future.
Disabled (0s) by defaultPlease note enable
this flag will reject samples in the future of
receive local NTP time + configured duration
due to clock skew in remote write clients.
--tsdb.wal-compression Compress the tsdb WAL.
--version Show application version.
Found a typo, inconsistency or missing information in our docs? Help us to improve Thanos documentation by proposing a fix on GitHub here ❤️