Query
get-configuration Retrieve the whole configuration.
get-degradations Get the list of fields for each layer, which fields and on which aggregation level, the field is degraded.
get-layers Get the list of layers, with the minimum granularity of each layer.
list-fields List all available fields and type descriptions.
query Request PVX data with the PVQL language.
query-cancel Cancel a PVQL Query.
Test
echo Echo back the passed arguments.
Other
cancel Cancel an API call.
get-api-version Return the API version.
Cancel an API call.
id
: The ID of the call to cancel.A boolean indicating if the cancellation succeeded.
=> cancel id="Query-DXGWL"
true
Echo back the passed arguments.
This is a test function, perfect to get a feel of the API.
=> echo name="Neptune" radius=24622
{"name": "Neptune", "radius": 24622}
Return the API version.
Retrieve the whole configuration.
An object with the following entries:
applications
applications_rules
nodes
version
zone_hierarchy
zone_rules
zones
This call will be deprecated once the configuration will be managed directly by the API.
Get the list of fields for each layer, which fields and on which aggregation level, the field is degraded.
{
<field>: {
<layer>: {
"degraded_on": [<aggregation_level>, ...]
},
...
},
...
}
Get the list of layers, with the minimum granularity of each layer.
{
<layer>: {
<min granularity>: <duration in seconds>
},
...
}
List all available fields and type descriptions.
An object of the form:
{
"fields": [{
"name": <string>,
"layers": [<string>],
"category": <string>,
"type": {...},
"description": {
"default": <string>,
"layer1" : <string>,
...
},
"title": {
"default": <string>,
"layer2" : <string>,
...
},
"short_title": {
"default": <string>,
"layer1" : <string>,
"layer2" : <string>,
...
},
"kind": "regular"
}, ...],
"types": {
<type>: {
"value_type": <string>,
"strict": <bool>,
"items": [{
"value": <id>,
"name" : <string>,
...
}, ...],
"default_format": <string>
},
...
}
}
On a more detailed example, if we want to retrieve the metadata from the field client.datasource.kind
:
{
"fields": [
{
"name": "client.datasource.kind",
"category": "key",
"type": {
"type": "pktsourcekind"
},
"layers": [
"citrix",
"citrix_channels",
"databases",
"dns",
"http",
"icmp",
"non_ip",
"other_ip",
"smb",
"tcp",
"tls",
"udp",
"voip"
],
"title": {
"default": "Client Packet Source Kind"
},
"short_title": {
"default": "Clt. PktSource Kind"
},
"description": {
"default": "The packet source type (pcap file, netflow, network iface, rpcapd) on which this traffic has been captured (client-side)"
},
"kind": "directed"
}, ...]
},
"types": {
[...]
"pktsourcekind": {
"value_type": "int",
"strict": false,
"items": [
{
"value": 0,
"name": "Missing"
},
{
"value": 1,
"name": "Network Interface"
},
{
"value": 2,
"name": "RPCAPD Remote Capture"
},
{
"value": 3,
"name": "PCAP File"
},
{
"value": 4,
"name": "NetFlow"
}
]
},
[...]
}
}
Request PVX data with the PVQL language.
expr
: The PVQL expression to execute.variables
: An optional object to map variables values used in the PVQL expression.format
: The format of the resulting data. Either plain
,
compact
or named
. Default to plain
. See more about data format
below.dry
: When set to true
, the query is only parsed and analyzed,
but it is not actually executed. This is convenient for interfaces
that need to validate syntax and to obtain meta information (which
columns will actually be requested for instance).See the PVQL topic for a detailed description of the language.
This function returns a series of chunks.
Each chunk contains one or more of the following entries:
info
: general information on the query.data
: portion of the result (the result can be split into multiple portions).meta
: metadata, closely related to the data
.More precisely, the chunk contains one of the following set of entries:
info
data
and meta
info
, data
and meta
info
entryThe info
entry contains important information, such as:
expr
: the requested expression.level
: the actual aggregation level (either explicitly specified by
the user, or automatically determined from the query).step
: the time span of a point in the result.timerange
: the actual time interval that was queried (which is adjusted
to align with the step).meta
: contains miscellaneous information:
meta.context
: all the extra variables made available for the query.meta.expectedRowCount
: a range of the number of rows to expect in
the result.meta.query_id
: PVQL Query ID provided by the SETTINGS query_id
clause.meta.sources
: the partitions of the database that were queried.meta.sort
: describes how to sort the data when split across multiple
chunks.meta.complexity
: contains complexity information.
meta.complexity.max_requested
: the maximum allowed complexity in the query
(null
if not bounded).meta.complexity.selected
: the selected complexity to run the query with.meta.complexity.thresholds
: the array of complexity thresholds to get more accurate results.The information in meta.sort
is structured as an array of items; each
item is in the form [key-or-value?, index, direction]
, where key-or-value?
is either 0 for a key, or 1 for a value, index
determines which key or value
to sort on (beware that the indices start at 0), and direction
is the sort
direction (either in ASCending or DESCending order).
we first sort on the first value expression (in descendant order), then on the first element of the group (is ascendant order).
"sort": [
["values", 0, "DESC"],
["key", 0, "ASC"]
]
data
entryThe data
entry contains the rows of data resulting from the execution
of the query.
Its structure depends of the format
argument.
plain
formatIn this format, the data are structured as an array of rows, where each row is made of two parts: the key, and the values.
For example, when querying traffic
, we obtain the following data:
{
"data": [{"key": [], "values": [{"value": 138229969750}]}]
}
Here, the data
entry contains a single row: [{"key": [], "values": [{"value": 138229969750}]}]
.
The key of the row ([]
), is empty because we are not grouping the values
(no BY
clause in the query).
On the other hand, the values part ([{"value": 138229969750}]
)
consists of one object whose value is the traffic.
It may happen that the field value is missing, for example because it was dropped during aggregation. In this case, the resulting object looks like
{"status": "missing"}
with a single attribute status
indicating why the value is missing
(missing
or invalid
).
Array values may be truncated for performance reasons. In this case,
the result objects come with both "value"
and "status": "truncated"
attributes.
Let’s consider another example, traffic BY server.ip[8] TOP 5
, which
results in the following chunk:
{
"data": [
{"key": [{"value": "10.0.0.0"}], "values": [{"value": 89930326910}]},
{"key": [{"value": "172.0.0.0"}], "values": [{"value": 31681460665}]},
{"key": [{"value": "192.0.0.0"}], "values": [{"value": 7825192006}]},
{"key": [{"value": "\u00f8"}], "values": [{"status": "empty"}]},
{"key": [{"value": "55.0.0.0"}], "values": [{"value": 62848673}]}
]
}
Due to grouping by server.ip[8]
, each row has a subnet in its key
portion.
compact
formatIn this format, the data are structured accordingly to the query.
When querying a single value, such as traffic
, the data is
represented by just this value:
{
"data": {"value": 138229969750}
}
When querying multiples values, with no groupment, such as traffic,
pdus
, the data is a simple array:
{
"data": [{"value": 138229969750}, {"value": 993478670}]
}
For queries with groupment, such as traffic BY server.ip[8] TOP 5
,
the format is close to the plain
format except that key and values
are grouped:
{
"data": [
[{"value": "10.0.0.0"}, {"value": 89930326910}],
[{"value": "172.0.0.0"}, {"value": 31681460665}],
[{"value": "192.0.0.0"}, {"value": 7825192006}],
[{"value": "\u00f8"}, {"status": "empty"}],
[{"value": "55.0.0.0"}, {"value": 62848673}]
]
}
named
formatIn this format, each row of the result is an object.
For example, for a query such as traffic BY server.ip[8] FROM transport TOP 5
, this
will return:
{
"data": [
{"<expr>": {"value": "10.0.0.0"}, "traffic": {"value": 89930326910}},
{"<expr>": {"value": "172.0.0.0"}, "traffic": {"value": 31681460665}},
{"<expr>": {"value": "192.0.0.0"}, "traffic": {"value": 7825192006}},
{"<expr>": {"value": "\u00f8"}, "traffic": {"status": "empty"}},
{"<expr>": {"value": "55.0.0.0"}, "traffic": {"value": 62848673}}
]
}
meta
entryThis entry is directly linked to the data
, i.e. there will be no
meta
entry unless there is a data
one.
It contains the following information:
sorted
: whether the rows in data
are sorted or not.elapsed
: the duration for building this chunk.accumulatedTime
: the total duration since the start of the query.emptyData
: whether there was actual data for the requested period
of time. If true
, when using compact
mode, a null
value
should be interpreted as the absence of data.totalRowsToRead
: total number of rows read by ClickHouse.Here is an example.
{
"meta": {
"accumulatedTime": 0.02259683609008789,
"elapsed": 0.017426252365112305,
"sorted": true,
"totalRowsToRead": 554457
}
}
Cancel a PVQL Query.
query_id
: PVQL Query ID.Nothing.