vtc man page

VTC — Varnish Test Case Syntax


This document describes the syntax used by Varnish Test Cases files (.vtc). A vtc file describe a scenario with different scripted HTTP-talking entities, and generally one or more Varnish instances to test.


A vtc file will be read word after word, with very little tokenization, meaning a syntax error won't be detected until the test actually reach the relevant action in the test.

A parsing error will most of the time result in an assert being triggered. If this happens, please refer yourself to the related source file and line number. However, this guide should help you avoid the most common mistakes.

Words and strings

The parser splits words by detecting whitespace characters and a string is a word, or a series of words on the same line enclosed by double-quotes ("..."), or, for multi-line strings, enclosed in curly brackets ({...}).


The leading whitespaces of lines are ignored. Empty lines (or ones consisting only of whitespaces) are ignored too, as are the lines starting with "#" that are comments.

Lines and commands

Test files take at most one command per line, with the first word of the line being the command and the following ones being its arguments. To continue over to a new line without breaking the argument string, you can escape the newline character (n) with a backslash ().



NOTE: this can be used from the top-level as well as from client and server specifications.

Barriers allows you to synchronize different threads to make sure events occur in the right order. It's even possible to use them in VCL.

First, it's necessary to declare the barrier:

barrier bNAME TYPE NUMBER [-cyclic]

With the arguments being:

this is the name of the barrier, used to identify it when you'll create sync points. It must start with 'b'.
it can be "cond" (mutex) or "sock" (socket) and sets internal behavior. If you don't need VCL synchronization, use cond.
number of sync point needed to go through the barrier.
if present, the barrier will reset itself and be ready for another round once gotten through.

Then, to add a sync point:

barrier bNAME sync

This will block the parent thread until the number of sync points for bNAME reaches the NUMBER given in the barrier declaration.

If you wish to synchronize the VCL, you need to declare a "sock" barrier. This will emit a macro definition named "bNAME_sock" that you can use in VCL (after importing the debug vmod):


This function returns 0 if everything went well and is the equivalent of barrier bNAME sync at the VTC top-level.


Client and server threads are fake HTTP entities used to test your Varnish and VCL. They take any number of arguments, and the one that are not recognized, assuming they don't start with '-', are treated as specifications, laying out the actions to undertake:

client cNAME [...]
server sNAME [...]

Clients and server are identified by a string that's the first argument, clients' names start with 'c' and servers' names start with 's'.

As the client and server commands share a good deal of arguments and specification actions, they are grouped in this single section, specific items will be explicitly marked as such.


Start the thread in background, processing the last given specification.
Block until the thread finishes.
-run (client only)
Equivalent to "-start -wait".
repeat NUMBER
Instead of processing the specification only once, do it NUMBER times.
-break (server only)
Stop the server.
-listen STRING (server only)
Dictate the listening socket for the server. STRING is of the form "IP PORT".
-connect STRING (client only)
Indicate the server to connect to. STRING is also of the form "IP PORT".
-dispatch (server only, s0 only)
Normally, to keep things simple, server threads only handle one connection at a time, but the -dispatch switch allows to accept any number of connection and handle them following the given spec.

However, -dispatch is only allowed for the server name "s0".
-proxy1 STRING (client only)
Use the PROXY protocol version 1 for this connection. STRING is of the form "CLIENTIP:PORT SERVERIP:PORT".
-proxy2 STRING (client only)
Use the PROXY protocol version 2 for this connection. STRING is of the form "CLIENTIP:PORT SERVERIP:PORT".

Macros and automatic behaviour

To make things easier in the general case, clients will connect by default to the first Varnish server declared and the -vcl+backend switch of the varnish command will add all the declared servers as backends.

Be careful though, servers will by default listen to the IP and will pick a random port, and publish 3 macros: sNAME_addr, sNAME_port and sNAME_sock, but only once they are started. For varnishtest to create the vcl with the correct values, the server must be started when you use -vcl+backend.


It's a string, either double-quoted "like this", but most of the time enclosed in curly brackets, allowing multilining. Write a command per line in it, empty line are ignored, and long line can be wrapped by using a backslash. For example:

client c1 {
    txreq -url /foo \
          -hdr "bar: baz"

} -run

accept (server)

Close the potential current connection, and accept a new one. Note that this new connection is H/1.

accept (server only)
Close the active connection (if any) and accept a new one.
Same as for the top-level barrier
chunked STRING
Send STRING as chunked encoding.
chunkedlen NUMBER
Do as chunked except that varnishtest will generate the string for you, with a length of NUMBER characters.

close (server)

Close the connection. Not that if operating in H/2 mode, no extra (GOAWAY) frame is sent, it's simply a TCP close.

close (server)

Close the connection. Not that if operating in H/2 mode, no extra (GOAWAY) frame is sent, it's simply a TCP close.

Same as for the top-level delay.
Same as for the top-level delay.

Test if "STRING1 OP STRING2" is true, and if not, fails the test. OP can be ==, <, <=, >, >= when STRING1 and STRING2 represent numbers in which case it's an order operator. If STRING1 and STRING2 are meant as strings OP is a matching operator, either == (exact match) or ~ (regex match).

varnishtet will first try to resolve STRING1 and STRING2 by looking if they have special meanings, in which case, the resolved value is use for the test. Note that this value can be a string representing a number, allowing for tests such as:

expect req.http.x-num > 2

Here's the list of recognized strings, most should be obvious as they either match VCL logic, or the txreq/txresp options:


expect_close (server)

Reads from the connection, expecting nothing to read but an EOF.

expect_close (server)

Reads from the connection, expecting nothing to read but an EOF.

Control whether a failure of this entity should stop the test.
Process STRING as a specification, NUMBER times.
Read NUMBER bytes from the connection.
Receive an HTTP chunk

rxpri (server)

Receive a preface, and if it matches, sets the server to H/2, aborts otherwise.

rxreq (server only)
Receive and parse a request's headers and body.
rxreqbody (server only)
Receive a request's body.
Receive and parse a request's headers (but not the body).
Receive and parse a request's headers (but not the body).
rxresp [-no_obj] (client only)
Receive and parse a response's headers and body. If -no_obj is present, only get the headers.
Receive a response's body.
Push STRING on the connection.
Write STRING on the socket NUMBER times.
send_urgent STRING
Send string as TCP OOB urgent data. You will never need this.
sendhex STRING
Send bytes as described by STRING. STRING should consist of hex pairs possibly separated by whitespace or newlines. For example: "0F EE a5 3df2".
settings -dectbl INT
Force internal H/2 settings to certain values. Currently only support setting the decoding table size.
timeout NUMBER
Set the TCP timeout for this entity.

txpri (client)

Send an H/2 preface ("PRI * HTTP/2.0\r\n\r\nSM\r\n\r\n") and set client to H/2.

txreq|txresp [...]

Send a minimal request or response, but overload it if necessary.

txreq is client-specific and txresp is server-specific.

The only thing different between a request and a response, apart from who can send them is that the first line (request line vs status line), so all the options are prety much the same.

-req STRING (txreq only)
What method to use (default: "GET").
-url STRING (txreq only)
What location to use (default "/").
-proto STRING
What protocol use in the status line. (default: "HTTP/1.1").
-status NUMBER (txresp only)
What status code to return (default 200).
-msg STRING (txresp only)
What message to put in the status line (default: "OK").

These three switches can appear in any order but must come before the following ones.

Don't include a Content-Length header in the response.
Add STRING as a header, it must follow this format: "name: value". It can be called multiple times.

You can then use the arguments related to the body:

-body STRING
Input STRING as body.
-bodylen NUMBER
Generate and input a body that is NUMBER bytes-long.
-gziplevel NUMBER
Set the gzip level (call it before any of the other gzip switches).
-gzipresidual NUMBER
Add extra gzip bits. You should never need it.
-gzipbody STRING
Zip STRING and send it as body.
-gziplen NUMBER
Combine -body and -gzipbody: create a body of length NUMBER, zip it and send as body.


H/2 introduces the concept of streams, and these come with their own specification, and as it's quite big, have bee move to their own chapter.


This is very similar to the the shell command, except it takes a first string as argument before the command:

err_shell "foo" "echo foo"

err_shell expect the shell command to fail AND stdout to match the string, failing the test case otherwise.


Test that the required feature(s) for a test are available, and skip the test otherwise. feature takes any number of arguments from this list:

The SO_RCVTIMEO socket option is working
The environment is 64 bits
The environment is not OSX
DNS lookups are working
varnishtest has been started with '-i'
varnishtest has been invoked by the root user
The varnish user is present
The vcache user is present
The varnish group is present
cmd <command-line>
A command line that should execute with a zero exit status


Reads the VSL and looks for records matching a given specification. It will process records trying to match the first pattern, and when done, will continue processing, trying to match the following pattern. If a pattern isn't matched, the test will fail.

logexpect threads are declared this way:

logexpect lNAME -v <id> [-g <grouping>] [-d 0|1] [-q query] \
        [vsl arguments] {
                expect <skip> <vxid> <tag> <regex>
                expect <skip> <vxid> <tag> <regex>
        } [-start|-wait]

And once declared, you can start them, or wait on them:

logexpect lNAME <-start|-wait>


Name the logexpect thread, it must start with 'l'.
-v id
Specify the varnish instance to use (most of the time, id=v1).
-g <session|request|vxid|raw
Decide how records are grouped, see -g in man varnishlog for more information.
-d <0|1>
Start processing log records at the head of the log instead of the tail.
-q query
Filter records using a query expression, see man vsl-query for more information.
Start the logexpect thread in the background.
Wait for the logexpect thread to finish

VSL arguments (similar to the varnishlog options):

Process only backend/client records.
Use caseless regex
-i <taglist>
Include tags
-I <[taglist:]regex>
Include by regex
-T <seconds>
Transaction end timeout

And the arguments of the specifications lines are:

skip: [uint|*]
Max number of record to skip
vxid: [uint|*|=]
vxid to match
tag: [tagname|*|=]
Tag to match against
regular expression to match against (optional) ('*' is anything, '=' is the value of the last matched record)


Pass the string given as argument to a shell. If you have multiple commands to run, you can use curly barces to describe a multi-lines script, eg:

shell {
        echo begin
        cat /etc/fstab
        echo end

The vtc will fail if the return code of the shell is not 0.


(note: this section is at the top-level for easier navigation, but it's part of the client/server specification)

Streams map roughly to a request in H/2, a request is sent on stream N, the response too, then the stream is discarded. The main exception is the first stream, 0, that serves as coordinator.

Stream syntax follow the client/server one:

stream ID [SPEC] [ACTION]

ID is the H/2 stream number, while SPEC describes what will be done in that stream.

Note that, when parsing a stream action, if the entity isn't operating in H/2 mode, these spec is ran before:

txpri/rxpri # client/server
stream 0 {
    txsettings -ack
    expect settings.ack == true
} -run

And H/2 mode is then activated before parsing the specification.


Run the specification in a thread, giving back control immediately.
Wait for the started thread to finish running the spec.
equivalent to calling -start then -wait.


The specification of a stream follows the exact same rules as one for a client or a server.

txreq, txresp, txcont, txpush

These four commands are about sending headers. txreq, txresp will send HEADER frames, txcont will send CONTINUATION frames, and txpush PUSH frames. The only difference between txreq and txresp are the default headers set by each of them.

Do not add default headers. Useful to avoid duplicates when sending default headers using -hdr, -idxHdr and -litIdxHdr.
-status INT (txresp)
Set the :status pseudo-header.
-url STRING (txreq, txpush)
Set the :path pseudo-header.
-req STRING (txreq, txpush)
Set the :method pseudo-header.
-scheme STRING (txreq, txpush)
Set the :scheme pseudo-header.
Insert a header, STRING1 being the name, and STRING2 the value.
-idxHdr INT
Insert an indexed header, using INT as index.
-litIdxHdr inc|not|never INT huf|plain STRING
Insert an literal, indexed header. The first argument specify if the header should be added to the table, shouldn't, or mustn't be compressed if/when retransmitted.

INT is the idex of the header name to use.

The third argument informs about the Huffman encoding: yes (huf) or no (plain).

The last term is the literal value of the header.
-litHdr inc|not|never huf|plain STRING1 huf|plain STRING2
Insert a literal header, with the same first argument as -litIdxHdr.

The second and third terms tell what the name of the header is and if it should be Huffman-encoded, while the last two do the same regarding the value.
-body STRING (txreq, txresp)
Specify a body, effectively putting STRING into a DATA frame after the HEADER frame is sent.
-bodylen INT (txreq, txresp)
Do the same thing as -body but generate an string of INT length for you.
-nostrend (txreq, txresp)
Don't set the END_STREAM flag automatically, making the peer expect a body after the headers.
Don't set the END_HEADERS flag automatically, making the peer expect more HEADER frames.
-dep INT (txreq, txresp)
Tell the peer that this content depends on the stream with the INT id.
-ex (txreq, txresp)
Make the dependency exclusive (-dep is still needed).
-weight (txreq, txresp)
Set the weight for the dependency.
-promised INT (txpush)
The id of the promised stream.
-pad STRING / -padlen INT (txreq, txresp, txpush)
Add string as padding to the frame, either the one you provided with -pad, or one that is generated for you, of length INT is -padlen case.


By default, data frames are empty. The receiving end will know the whole body has been delivered thanks to the END_STREAM flag set in the last DATA frame, and txdata automatically set it.

-data STRING
Data to be embedded into the frame.
-datalen INT
Generate and INT-bytes long string to be sent in the frame.
-pad STRING / -padlen INT
Add string as padding to the frame, either the one you provided with -pad, or one that is generated for you, of length INT is -padlen case.
Don't set the END_STREAM flag, allowing to send more data on this stream.

rxreq, rxresp

These are two convenience functions to receive headers and body of an incoming request or response. The only difference is that rxreq can only be by a server, and rxresp by a client.


This works like rxhdrs, expecting a PUSH frame and then zero or more CONTINUATION frames.

Keep waiting for CONTINUATION frames until END_HEADERS flag is seen.
-some INT
Retrieve INT - 1 CONTINUATION frames after the PUSH frame.


rxhdrs will expect one HEADER frame, then, depending on the arguments, zero or more CONTINUATION frame.

Keep waiting for CONTINUATION frames until END_HEADERS flag is seen.
-some INT
Retrieve INT - 1 CONTINUATION frames after the HEADER frame.


Receiving data is done using the rxdata keywords and will retrieve one DATA frame, if you wish to receive more, you can use these two convenience arguments:

keep waiting for DATA frame until one sets the END_STREAM flag
-some INT
retrieve INT DATA frames.


Take a float as argument and sleep for that number of seconds.

Receive a frame, any frame.


Push bytes directly on the wire. sendhex takes exactly one argument: a string describing the bytes, in hex notation, will possible whitespaces between them. Here's an example:

sendhex "00 00 08 00 0900       8d"


Receive a GOAWAY frame


Possible options include:

set the error code to eplain the termination. The second argument can be a integer or the string version of the error code as found in rfc7540#7.
-laststream INT
the id of the "highest-numbered stream identifier for which the sender of the GOAWAY frame might have taken some action on or might yet take action on".
specify the debug data, if any to append to the frame.


Receive a PING frame


Send PING frame.

-data STRING
specify the payload of the frame, with STRING being an 8-char string.
set the ACK flag.


Receive a PRIORITY frame


Send a PRIORITY frame

-stream INT
indicate the id of the stream the sender stream depends on.
the dependency should be made exclusive (only this streams depends on the parent stream).
-weight INT
an 8-bits integer is used to balance priority between streams depending on the same streams.


Receive a RST_STREAM frame


Send a RST_STREAM frame. By default, txrst will send a 0 error code (NO_ERROR).

Sets the error code to be sent. The argument can be an integer or a string describing the error, such as NO_ERROR, or CANCEL (see rfc7540#11.4 for more strings).


Receive a SETTINGS frame


SETTINGS frames must be acknowledge, arguments are as follow (most of them are from rfc7540#6.5.2):

-hdrtbl INT
headers table size
-push BOOL
whether push frames are accepted or not
-maxstreams INT
maximum concurrent streams allowed
-winsize INT
sender's initial window size
-framesize INT
largest frame size authorized
-hdrsize INT
maximum size of the header list authorized
set the ack bit


Receive a WINDOW_UPDATE frame


Transmit a WINDOW_UPDATE frame, increasing the amount of credit of the connection (from stream 0) or of the stream (any other stream).

-size INT
give INT credits to the peer.


expect in stream works as it does in client or server, except that the elements compared will be different.

Most of these elements will be frame specific, meaning that the last frame received on that stream must of the correct type.

Here the list of keywords you can look at.


Define and interact with varnish instances.

To define a Varnish server, you'll use this syntax:

varnish vNAME [-arg STRING] [-vcl STRING] [-vcl+backend STRING]
        [-errvcl STRING STRING] [-jail STRING] [-proto PROXY]

The first varnish vNAME invocation will start the varnishd master process in the background, waiting for the -start switch to actually start the child.


Identify the Varnish server with a string, it must starts with 'v'.
Pass an argument to varnishd, for example "-h simple_list".
Specify the VCL to load on this Varnish instance. You'll probably want to use multi-lines strings for this ({...}).
-vcl+backend STRING
Do the exact same thing as -vcl, but adds the definition block of known backends (ie. already defined).
Load STRING2 as VCL, expecting it to fail, and Varnish to send an error string matching STRING2
-jail STRING
Look at man varnishd (-j) for more information.
-proto PROXY
Have Varnish use the proxy protocol. Note that PROXY here is the actual string.

You can decide to start the Varnish instance and/or wait for several events:

varnish vNAME [-start] [-wait] [-wait-running] [-wait-stopped]
Start the child process.
Stop the child process.
Set the VCL syntax level (default: 4.0)
Wait for that instance to terminate.
Wait for the Varnish child process to be started.
Wait for the Varnish child process to stop.
Once Varnish is stopped, clean everything after it. This is only used in one test and you should never need it.

Once Varnish is started, you can talk to it (as you would through varnishadm) with these additional switches:

varnish vNAME [-cli STRING] [-cliok STRING] [-clierr STRING]
              [-expect STRING OP NUMBER]
All three of these will send STRING to the CLI, the only difference is what they expect the return code to be. -cli doesn't expect anything, -cliok expects 200 and -clierr expects STATUS

Look into the VSM and make sure the counter identified by STRING has a correct value. OP can be ==, >, >=, <, <=. For example:

varnish v1 -expect SMA.s1.g_space > 1000000
Dump VSC counters matching PATTERN. The PATTERN is a 'glob' style pattern (ie: fnmatch(3)) as used in shell filename expansion. To see all counters use pattern "*", to see all counters about requests use "req".


This should be the first command in your vtc as it will identify the test case with a short yet descriptive sentence. It takes exactly one argument, a string, eg:

varnishtest "Check that varnishtest is actually a valid command"

It will also print that string in the log.


This document has been written by Guillaume Quintard.

See Also


Referenced By


Explore man page connections for vtc(7).