SearchVarnish Enterprise
The probe proxy VCL allows users to redirect probes to a new location or enables
probes to first pass through VCL. This gives probes all of the flexibility
that VCL provides including routing, caching and collapsing.
There are two main modes of operations. The first is an override of a singular backend. This mode redirects a probe from the original backend to a different backend. All of the facilities of a probe work as a traditional probe would. The second option is a global override that will have all probes go through Varnish first then go to the original backend or a secondary backend. Using a global override gives the probe request access to the full set of VCL features.
This is a VCL API that is enabled when a VCL includes probe-proxy.vcl.
Each probe request will first be routed through Varnish and then to the
original backend or to a gateway backend. By default, probes are cached
according to the URL of the request and the Backend. Each request has a
set of headers to describe the probe and backend it is attached to.
Additional VCL to manage probes can be added to sub vcl_recv and sub vcl_deliver
with the call_recv and call_deliver options. Optionally a security token
can be added with the set_token() function. When used each request will
check if it has the token. This mode requires a listening address that isn’t TLS or a proxy.
To use Global Override you must include the probe-proxy.vcl, this must be
included above other VCL includes such as VHA. To edit other settings see below.
include "probe-proxy.vcl";
Each request gets a set of request headers describing the probe and backend that it is attached to.
VPP-probe-proxy - true when the request is from the Global Override state.VPP-token - When set_token() is called, it contains the security token to check.VPP-backend-name - The name of the backend the probe is attached to.VPP-timeout - The .timeout set in the probe’s definition. Used for the connect timeout and first byte timeout of the request.VPP-interval - The .interval set in the probe’s definition. Used to set the TTL of the request.VPP-exp_status - The .exp_status set in the probe’s definition.VPP-window - The .window set in the probe’s definition.VPP-threshold - The .threshold set in the probe’s definition.VPP-initial - The .initial set in the probe’s definition.All settings are optional and can only be applied in sub vcl_init.
probe_proxy.set_token(STRING token)
Add a security token to validate a probe request.
probe_proxy_gateway.add_backend(BACKEND be)
Send all probe requests to backend be. If this backend is not available
the request will then go to the original backend.
probe_proxy_opts.set("per_host", "false");
coalesce probe requests per host header instead of per backend. This enables a per director probe instead of the default per backend probe.
probe_proxy_opts.set("call_recv", "false");
Use additional VCL in sub vcl_recv. The set of headers defined above are
available for information about the request. This VCL must be in a
.is_probe() conditional statement. Must return hash afterwards.
See Add VCL to probes for example of how this VCL should look.
probe_proxy_opts.set("call_deliver", "false");
Use additional VCL in sub vcl_deliver. The set of headers defined above are
available for information about the request. This VCL must be in a .is_probe()
conditional statement. Must return deliver afterwards. See Add VCL to probes
for example of how this VCL should look.
probe_proxy_opts.set("call_backend_response", "false");
Use additional VCL in sub vcl_backend_response. The set of headers defined above are
available for information about the request. This VCL must be in a .is_probe()
conditional statement. Must return deliver afterwards. See Add VCL to probes
for example of how this VCL should look.
probe_proxy_opts.set("retries", "0");
How many times should the request retry the gateway before falling back to the original backend.
This sends health_check to new_probe_location instead of default
for default’s health check.
backend default {
.host = "host";
.port = "80";
}
backend new_probe_location {
.host = "host2";
.port = "82";
}
probe health_check {
.url = "/1";
}
import probe_proxy;
sub vcl_init {
probe_proxy.override(default,
health_check,
new_probe_location);
}
Global and per backend VCL can be used for probe requests.
probe health_check {
.url = "/";
}
backend host1 {
.host = "host1";
.port = "80";
.probe = health_check;
}
backend host2 {
.host = "host2";
.port = "80";
.probe = health_check;
}
include "probe-proxy.vcl";
sub vcl_init {
probe_proxy_opts.set("call_recv", "true");
probe_proxy_opts.set("call_deliver", "true");
probe_proxy_opts.set("call_backend_response", "true");
}
sub vcl_recv {
if (probe_proxy.is_probe()) {
set req.http.user-agent = "Varnish Probe";
# Only add an header to probes for host2
if (req.http.VPP-backend-name == "host2") {
set req.http.Authorization = "basic base64string";
}
return (hash);
}
}
sub vcl_deliver {
if (probe_proxy.is_probe()) {
# Consider this range of response codes valid
if (resp.status > 200 && resp.status < 300) {
set resp.status = 200;
}
return (deliver);
}
}
sub vcl_backend_response {
if (probe_proxy.is_probe()) {
# Tolerate 404 from host2
if (beresp.backend.name == "host2" && beresp.status == 404) {
set beresp.status = 200;
}
return (deliver);
}
}
When a director creates dynamic backends the amount of probe requests to a service will increase by the number of backends created. This feature will coalesce a probe request to a director.
import goto;
include "probe-proxy.vcl";
probe health_check {
.url = "/";
}
sub vcl_init {
probe_proxy_opts.set("per_host", "true");
new dyn_dir = goto.dns_director("dynamic_host",
probe = health_check);
}
VOID override(BACKEND be, PROBE p, BACKEND new_location)
Override the destination of a probe. be will use new_location as it’s probe destination.
Can only be called in sub vcl_init.
Arguments:
be accepts type BACKEND
p accepts type PROBE
new_location accepts type BACKEND
Type: Function
Returns: None
Restricted to: vcl_init
VOID global_override(BACKEND be)
Override the probe location of all backends to backend be.
Arguments:
be accepts type BACKENDType: Function
Returns: None
Restricted to: vcl_init
VOID set_token(STRING token)
Add a token header to the global override header set.
Arguments:
token accepts type STRINGType: Function
Returns: None
Restricted to: vcl_init
BOOL is_probe()
Check if a request is from a global override.
Arguments: None
Type: Function
Returns: Bool
Restricted to: client, backend
BACKEND backend([STRING name])
Return a backend from a given name. Default name is from VPP-backend-name header.
Arguments:
name accepts type STRINGType: Function
Returns: Backend
Restricted to: client, backend
BACKEND self([STRING name])
Generate a backend from the first valid listening address of Varnish.
Arguments:
name accepts type STRINGType: Function
Returns: Backend
VOID skip_health_check()
Enable bo flag to skip health check on a backend request.
Arguments: None
Type: Function
Returns: None
Restricted to: vcl_backend_fetch
VOID force_fresh()
Enable bo flag to force a fresh connection for a backend request.
Arguments: None
Type: Function
Returns: None
Restricted to: vcl_backend_fetch
DURATION timeout()
Convert the VPP-timeout header to duration. If not defined return default timeout of 2.
Arguments: None
Type: Function
Returns: Duration
Restricted to: client, backend
DURATION interval()
Convert the VPP-interval header to duration. If not defined return default timeout of 5.
Arguments: None
Type: Function
Returns: Duration
Restricted to: client, backend
The probe_proxy VMOD is available in Varnish Enterprise version 6.0.8r2 and later.