Examples

This section will highlight some of the simple use cases where the broadcaster can be used. While it was meant as a generic http distributor, its main purpose resides around the invalidation of cached content.

The API path will, from version 1.2.4, support versioning in the form of /api/v1. It will still support /api/ for now, which points to version v1.

The examples below uses the following nodes.conf:

[GroupA]
Cache1=http://example.com:81
Cache2=http://example.com:82
Cache3=http://example.com:83

[GroupB]
Cache4=http://example.com:84
Cache5=http://example.com:85

[GroupC]
Cache6=http://example.com:86
Cache7=http://example.com:87
Cache8=http://example.com:88

Invalidation examples

For each case (purge, ban, ykey), these examples assumes the Varnish nodes are configured like described in the invalidation tutorial.

Purge

A purge will invalidate all objects with the same hash as the invalidation requests, so make sure to specify the host header:

# Purge the /something/to/purge URL from all the nodes
curl http://localhost:8088/something/to/purge -X PURGE -H "host: my.domain.com"

# Only target the GroupA nodes, but connect the broadcaster via HTTPS
curl https://localhost:8088/something/to/purge -X PURGE -H "host: my.domain.com" -H "X-Broadcast-Group: GroupA" --cacert server.crt

Ban

For banning, we have access to regular expressions, allowing to cache multiple subdomains and entire subtrees.

# Ban all objects under /foo/
curl -is http://localhost:8088/ -X BAN -H "ban-url: ^/foo/"

# Ban everything in a *.foo.com subdomain
curl -is http://localhost:8088/ -X BAN -H "ban-host: \.foo\.com$"

# Ban all the content in foo.com/static/ from the `GroupB` nodes
curl -is http://localhost:8088/ -X BAN -H "ban-host: ^foo\.com$" -H "ban-url: ^/static/" -H "X-Broadcast-Group: GroupA"

YKey

Here, we just need the id that needs to be invalidated.

# Invalidate all ykey objects marked with "a1b2c3d4f5g6"
curl -is http://localhost:8088/ -H "ykey-purge: a1b2c3d4f5g6"

X-Broadcast-Random Header

The broadcaster can also act as a load balancer and pick only one node, at random, out of a group, using the X-Broadcast-Group.

If a group is treated “randomly”, the broadcaster will pick only one node from it, but if it can’t get an HTTP response out of it, it will cycle through the group until it get a response.

Note: the HTTP code of the response isn’t taken into account, only whether or not a response exist.

The header X-Broadcast-Group has precendence over X-Broadcast-Random, so if a group appears in both, the group will be treated fully.

As an example you can send a request to one node in each group:

curl -is https://localhost:8088/endpoint/to/trigger -H "X-Broadcast-Random: *"

With a possible response being:

HTTP/1.1 200 OK
Content-Type: application/json
Date: Mon, 11 Dec 2017 12:34:33 GMT
Content-Length: 120

{
  "done": true,
  "method": "GET",
  "uri": "/something/to/fetch",
  "ts": 1512995646,
  "rate": 100,
  "nodes": {
    "Cache2": 200,
    "Cache5": 200,
    "Cache8": 200,
  }
}

But you can also send the request to all the GroupA nodes and pick just one out of GroupB and GroupC

curl -is https://localhost:8088/endpoint/to/trigger -H "X-Broadcast-Random: *" -H "X-Broadcast-Group: GroupA"

Giving us:

HTTP/1.1 200 OK
Content-Type: application/json
Date: Mon, 11 Dec 2017 12:34:33 GMT
Content-Length: 101

{
  "done": true,
  "method": "GET",
  "uri": "/something/to/fetch",
  "ts": 1512995646,
  "rate": 100,
  "nodes": {
    "Cache1": 200,
    "Cache2": 200,
    "Cache3": 200,
    "Cache5": 200,
  }
}

X-Broadcast-InOrder Header

A multi-layer cache setup will have some caches closer to the origin and other caches closer to the user. To ensure data consistency, the content must be purged first from the layer closest to the origin, then progress downstream, all the way to the edge servers. This means that groups must be purged in order instead of all at once, which is what this header enables.

Assuming that in the example nodes.conf: - GroupA is the origin shield, right in front of the origin, - GroupB is a storage tier using GroupA as backend, - and GroupC is the edge tier serving the users, and using GroupB as backend, then we can purge them all, in the right order with:

$ curl -is localhost:8088/foo -X PURGE -H "host: my.domain.com" -H "X-Broadcast-Group: GroupA GroupB GroupC" -H "X-Broadcast-InOrder: true"

Note the default group order is taken from the config file, from top to bottom. Taking advantage of this we can omit the X-Broadcast-Group header and get the same purge order of the previous command.

$ curl -is localhost:8088/foo -X PURGE -H "host: my.domain.com" -H "X-Broadcast-InOrder: true"

Response samples

Output sample in synchronous mode:

HTTP/1.1 200 OK
Content-Type: application/json
Date: Mon, 11 Dec 2017 12:34:33 GMT
Content-Length: 101

{
  "done": true,
  "method": "PURGE",
  "uri": "/foo/bar",
  "ts": 1512995646,
  "rate": 25,
  "nodes": {
    "Cache1": 503,
    "Cache2": 500,
    "Cache3": 503,
    "Cache4": 200
  }
}

Output sample in async mode:

HTTP/1.1 202 Accepted
X-Job-Id: 570204483
Date: Mon, 11 Dec 2017 12:20:59 GMT
Content-Length: 0
Content-Type: text/plain; charset=utf-8

When in async mode, the broadcaster returns imediatelly to the client with the status code of 202 and a X-Job-Id header. Use the value of this header in the /api/status endpoint to check whether the invalidation request has finished.

Configuration query

curl http://localhost:8089/api/configuration

HTTP/1.1 200 OK
Access-Control-Allow-Origin: *
Access-Control-Expose-Headers: X-Job-Id
Content-Type: application/json
Date: Wed, 12 Dec 2017 09:40:59 GMT
Content-Length: 313

{
    "Local": [
        {
            "address": "https://localhost:7081",
            "name": "beta"
        },
        {
            "address": "https://localhost:4443",
            "name": "charlie"
        },
        {
            "address": "https://localhost:6081",
            "name": "alpha"
        }
    ],
    "Remote": [
        {
            "address": "https://burlan.eu",
            "name": "second"
        },
        {
            "address": "https://apienginedemo.varnish-software.com",
            "name": "third"
        }
    ],
	"async": false
}

Ping query

curl -I http://localhost:8089/api/ping

HTTP/1.1 200 OK
Access-Control-Allow-Origin: *
Access-Control-Expose-Headers: X-Job-Id
Content-Length: 0
X-Pong: 1631603497
Date: Tue, 14 Sep 2021 07:11:37 GMT

All request status

curl -is http://localhost:8089/api/status

HTTP/1.1 200 OK
Content-Type: application/json
Date: Mon, 11 Dec 2017 12:42:29 GMT
Content-Length: 574

{
  "510158363": {
    "done": true,
    "method": "PURGE",
    "uri": "/foo/bar",
    "ts": 1512995673,
    "rate": 25,
    "nodes": {
      "Cache1": 503,
      "Cache2": 500,
      "Cache3": 503,
      "Cache4": 200
    }
  },
  "752412541": {
    "done": true,
    "method": "BAN",
    "uri": "/foo/bar",
    "ts": 1512995630,
    "rate": 25,
    "nodes": {
      "Cache1": 503,
      "Cache2": 500,
      "Cache3": 503,
      "Cache4": 200
    }
  },
  "2377945909": {
    "done": true,
    "method": "PURGE",
    "uri": "/foo/bar",    
    "ts": 1512995637,
    "rate": 25,
    "nodes": {
      "Cache1": 503,
      "Cache2": 500,
      "Cache3": 503,
      "Cache4": 200
    }
  }
}

Finished request status

curl -is http://localhost:8089/api/status?id=2377945909

HTTP/1.1 200 OK
Content-Type: application/json
Date: Mon, 11 Dec 2017 12:42:43 GMT
Content-Length: 101

{
    "done": true,
    "ts": 1512995637,
    "method": "PURGE",
    "uri": "/foo/bar",    
    "rate": 25,
    "nodes": {
      "Cache1": 503,
      "Cache2": 500,
      "Cache3": 503,
      "Cache4": 200
    }
}

Ongoing request status

curl -is http://localhost:8089/api/status?id=752412541

HTTP/1.1 202 Accepted
X-Job-Id: 570204483
Date: Mon, 11 Dec 2017 12:20:59 GMT
Content-Length: 0
Content-Type: text/plain; charset=utf-8

Statistics

The stats endpoint returns the statistics for broadcasts. It will return the structure shown below in JSON format. The statistics is grouped per configured groups and nodes. It keeps statistics for methods used, counter for responses from nodes and an aggregated TotalRequests per scope.

sinceRequests shows the time since the statstics was last reset.
started is the timestamp when the statistics was last reset.
totalBroadcasts is the number of broadcasts performed since the last reset.
requestsPerSec is number of requests per second to nodes since reset.

Fetch current statistics:

curl -is http://localhost:8089/api/stats

HTTP/1.1 200 OK
Access-Control-Allow-Origin: *
Access-Control-Expose-Headers: X-Job-Id
Content-Type: application/json
Date: Tue, 20 Oct 2020 07:17:43 GMT
Content-Length: 720

{
  "started": "2020-10-20T09:17:40.458600175+02:00",
  "totalBroadcasts": 1,
  "sinceReset": "2.725s",
  "groups": {
    "global.cluster": {
      "hosts": {
        "server1": {
          "address": "http://127.0.0.1:1234",
          "requests": {
            "GET": {
              "200": 1
            }
          }
        },
        "server2": {
          "address": "http://127.0.0.1:1235",
          "requests": {
            "GET": {
              "500": 1
            }
          }
        }
      },
      "totalRequests": {
        "GET": {
          "200": 1,
          "500": 1
        }
      }
    }
  },
  "totalRequests": {
    "GET": {
      "200": 1,
      "500": 1
    }
  },
  "requestsPerSec": 0

Reset statistics:

curl -is localhost:8089/api/v1/stats/reset

HTTP/1.1 200 OK
Access-Control-Allow-Origin: *
Access-Control-Expose-Headers: X-Job-Id
Content-Type: application/json
Date: Tue, 20 Oct 2020 07:21:28 GMT
Content-Length: 0

Status API

The status information for any invalidation request has the following structure:

Name	About
`method`	The HTTP method used for broadcasting.
`uri`	The URI to be broadcasted against the configured nodes.
`done`	Boolean telling whether the job batch has finished.
`ts`	Unix timestamp which tells the time when a specific invalidation request has been sent onto the working channel.
`rate`	Percentage which represents the quota of succesfull broadcasts. A broadcast is considered to have succeeded if its returned status code is `200`.
`nodes`	Dictionary representation of the nodes which have been broadcasted against along with their returned status codes.
`err`	If present, this tag will contain errors occured at broadcast time.