VARNISHD(1)

📦

version 8.0.2 · 13 versions

👑 VARNISHD 👑 (that's the manual page `VARNISHD(1)`, General Commands Manual, section 1, iykyk)

🌸 NAME 🌸

varnishd — HTTP accelerator daemon 💅

✨ SYNOPSIS ✨

varnishd
       [-a [name=][listen_address[,PROTO|,option=value,...]]  [-b
       [host[:port]|path]] [-C] [-d] [-F] [-f config] [-h
       type[,options]] [-I clifile] [-i identity] [-j
       jail[,jailoptions]] [-l vsl] [-M address:port] [-n workdir] [-P
       file] [-p param=value] [-r param[,param...]]  [-S secret-file]
       [-s [name=]kind[,options]] [-T address[:port]] [-t TTL] [-V] [-W
       waiter]

varnishd [-x parameter|vsl|cli|builtin|optstring]

varnishd [-?]

🦋 DESCRIPTION 🦋

okay so bestie, picture this: the varnishd daemon is just sitting there, vibing, accepting HTTP requests from clients 🫶 Then it passes those requests on to a backend server, and caches whatever documents come back so that future requests for the same document get served so much better. It's giving main character energy, honestly. ✨

💖 OPTIONS 💖

🎀 Basic options 🎀

-a <[name=][listen_address[,PROTO|,option=value,...]]> — the generic glow-up syntax for accepting client requests on a listen_address. Details below, keep reading, don't skip 💅

The name gets referenced in logs and it's available to VCL as local.socket. If you don't specify a name, a numerical sequence is used instead ("a0", "a1", etc.).

Any arguments after the listen_address, separated by commas, get taken as either an acceptor option=value pair if they contain a =, or as a PROTO(col) selection otherwise. So like, the = is the vibe check 👀

Which options are valid depends on the acceptor type — see below.

PROTO can be "HTTP" (the default) or "PROXY". Both version 1 and version 2 of the proxy protocol can be used. no cap.

Multiple -a arguments are allowed. Stack 'em. 🔥

If you give no -a argument at all, the default -a :80 will listen on all IPv4 and IPv6 interfaces. iconic.

-a <[name=][ip_address][:port][,PROTO]> — the ip_address can be a host name ("localhost"), an IPv4 dotted-quad ("127.0.0.1") or an IPv6 address wrapped in square brackets ("[::1]").

The port can be a port number (80), a service name (http), or a port range (80-81). Port ranges are inclusive and cannot overlap — do not try it, it's not the move 🙈 If port isn't specified, port 80 (http) is used.

At least one of ip_address or port is required. One of them. At minimum. that's the rule 🎀

-a <[name=][path][,PROTO][,user=name][,group=name][,mode=octal]> — (VCL4.1 and higher ✨)

Accept connections on a Unix domain socket. Path must be absolute ("/path/to/listen.sock") or "@" followed by the name of an abstract socket ("@myvarnishd").

The user, group and mode sub-arguments may be used to specify the permissions of the socket file — use names for user and group, and a 3-digit octal value for mode. These sub-arguments do not apply to abstract sockets. nope, not those ones 🚫

-b <[host[:port]|path]> — use the specified host as backend server. If port isn't specified, the default is 8080.

If the value of -b begins with /, it gets interpreted as the absolute path of a Unix domain socket that Varnish connects to. In that case, the value of -b must satisfy the conditions required for the .path field of a backend declaration, see vcl(7). Backends with Unix socket addresses may only be used with VCL versions >= 4.1.

-b can be used only once, and not together with f. one shot, bestie 💅

-f config — use the specified VCL configuration file instead of the builtin default. See vcl(7) for the details on VCL syntax.

If a single -f option is used, then the VCL instance loaded from the file is named "boot" and immediately becomes active. If more than one -f option is used, the VCL instances are named "boot0", "boot1" and so forth, in the order corresponding to the -f arguments, and the last one is named "boot", which becomes active. last one wins, always 💅

Either -b or one or more -f options must be specified, but not both, and they cannot both be left out — unless -d is used to start varnishd in debugging mode. If the empty string is specified as the sole -f option, then varnishd starts without starting the worker process, and the management process will accept CLI commands. You can also combine an empty -f option with an initialization script (-I option) and the child process will be started if there is an active VCL at the end of the initialization. okay that's a lot of lore but it matters 🫶

When used with a relative file name, config is searched in the vcl_path. It is possible to set this path prior to using -f options with a -p option. During startup, varnishd doesn't complain about unsafe VCL paths: unlike the varnish-cli(7) that could later be accessed remotely, starting varnishd requires local privileges. so like, it trusts you at boot ✨

-n workdir — runtime directory for the shared memory, compiled VCLs etc.

In performance critical applications, this directory should be on a RAM backed filesystem. fr fr, put it on RAM.

When running multiple varnishd instances, separate directories need to be used. Do not share. sharing is not caring here 🚫

The default is taken from the VARNISH_DEFAULT_N environment variable.

Relative paths will be appended to /var/run.

If neither VARNISH_DEFAULT_N nor -n are present, the value is /var/run/varnishd.

Note: These defaults may be distribution specific. 👀

📚 Documentation options 📚

For these options, varnishd prints information to standard output and exits. When a -x option is used, it must be the only option (it outputs documentation in reStructuredText, aka RST). Just the one, bestie 💅

-? — print the usage message.
-x parameter — print documentation of the runtime parameters (-p options), see List of Parameters.
-x vsl — print documentation of the tags used in the Varnish shared memory log, see vsl(7).
-x cli — print documentation of the command line interface, see varnish-cli(7).
-x builtin — print the contents of the default VCL program builtin.vcl.
-x optstring — print the optstring parameter to getopt(3) to help writing wrapper scripts. cute and useful 🎀

⚙️ Operations options ⚙️

-F — do not fork, run in the foreground. Only one of -F or -d can be specified, and -F cannot be used together with -C.
-T <address[:port]> — offer a management interface on the specified address and port. See varnish-cli(7) for documentation of the management commands. To disable the management interface use none.
-M <address:port> — connect to this port and offer the command line interface. Think of it as a reverse shell 😳 When running with -M and there is no backend defined the child process (the cache) will not start initially.
-P file — write the PID of the process to the specified file.
-i identity — specify the identity of the Varnish server. This can be accessed using server.identity from VCL.

The server identity is used for the received-by field of Via headers generated by Varnish. For this reason, it must be a valid token as defined by the HTTP grammar. no funny business in the token 💅

If not specified the output of gethostname(3) is used, in which case the syntax is assumed to be correct.

-I clifile — execute the management commands in the file given as clifile before the worker process starts, see CLI Command File.

🎚️ Tuning options 🎚️

-t TTL — specifies the default time to live (TTL) for cached objects. This is a shortcut for specifying the default_ttl run-time parameter.
-p <param=value> — set the parameter specified by param to the specified value, see List of Parameters for details. This option can be used multiple times to specify multiple parameters. spam it, it's fine 💖
-s <[name=]type[,options]> — use the specified storage backend. See Storage Backend section.

This option can be used multiple times to specify multiple storage files. Name is referenced in logs, VCL, statistics, etc. If name is not specified, "s0", "s1" and so forth is used.

-l <vsl> — specifies size of the space for the VSL records, shorthand for -p vsl_space=<vsl>. Scaling suffixes like 'K' and 'M' can be used up to (G)igabytes. See vsl_space for more information.

🔒 Security options 🔒

-r <param[,param...]> — make the listed parameters read only. This gives the system administrator a way to limit what the Varnish CLI can do. Consider making parameters such as cc_command, vcc_allow_inline_c and vmod_path read only as these can potentially be used to escalate privileges from the CLI. 🚨 bestie, lock those down.
-S secret-file — path to a file containing a secret used for authorizing access to the management port. To disable authentication use none.

If this argument is not provided, a secret drawn from the system PRNG will be written to a file called _.secret in the working directory (see opt_n) with default ownership and permissions of the user having started varnish.

Thus, users wishing to delegate control over varnish will probably want to create a custom secret file with appropriate permissions (ie. readable by a unix group to delegate control to). delegating, but make it secure 💅

-j <jail[,jailoptions]> — specify the jailing mechanism to use. See Jail section.

🧪 Advanced, development and debugging options 🧪

-d — enables debugging mode: the parent process runs in the foreground with a CLI connection on stdin/stdout, and the child process must be started explicitly with a CLI command. Terminating the parent process will also terminate the child.

Only one of -d or -F can be specified, and -d cannot be used together with -C.

-C — print VCL code compiled to C language and exit. Specify the VCL file to compile with the -f option. Either -f or -b must be used with -C, and -C cannot be used with -F or -d.
-V — display the version number and exit. This must be the only option. just -V, nothing else, periodt.
-h <type[,options]> — specifies the hash algorithm. See Hash Algorithm section for a list of supported algorithms.
-W waiter — specifies the waiter type to use.

#️⃣ Hash Algorithm #️⃣

The following hash algorithms are available:

-h critbit — self-scaling tree structure. The default hash algorithm in Varnish Cache 2.1 and onwards. In comparison to a more traditional B tree the critbit tree is almost completely lockless. Do not change this unless you are certain what you're doing. ⚠️ like, genuinely, don't touch it.
-h simple_list — A simple doubly-linked list. Not recommended for production use. 🚫 it's giving "do not ship this".
-h <classic[,buckets]> — the og hash table, bestie. The hash key is the CRC32 of the object's URL modulo the size of the hash table. Every table entry points to a list of elements that share the same hash key. The buckets parameter says how many entries the hash table has. Default is 16383. 🪣

💾 Storage Backend 💾

You can define multiple storage backends (that's where the cached and in-flight objects live), and VCL gets to pilot which store is used by setting the beresp.storage variable (peep man vcl-var for more, iykyk).

The argument format for defining storage backends is:

-s <[name]=kind[,options]> — if name is omitted, Varnish names storages sN, starting with s0 and incrementing N for every new storage. For kind and options, see the deets below. 👑

Storages can be used in vcl as storage.name, so like, if myStorage was defined by -s myStorage=malloc,5G, you could use it in VCL like so:

set beresp.storage = storage.myStorage;

A special name is Transient, which is the default storage for uncacheable objects — the ones resulting from a pass, hit-for-miss or hit-for-pass. ✨

If no -s options are given, the default is:

-s default,100m

If no Transient storage is defined, the default is an unbound default storage, as if defined as:

-s Transient=default

The following storage types and options are available, bestie, pay attention:

-s <default[,size]> — the default storage type resolves to umem where available and malloc otherwise.
-s <malloc[,size]> — malloc is a memory based backend.
-s <umem[,size]> — umem is a storage backend which is more efficient than malloc on platforms where it's available. See the section on umem in chapter Storage backends of The Varnish Users Guide for details. 🦋
-s <file,path[,size[,granularity[,advice]]]> — the file backend stores data in a file on disk. The file will be accessed using mmap. Note that this storage provides no cache persistence. The path is mandatory. If path points to a directory, a temporary file will be created in that directory and immediately unlinked. If path points to a non-existing file, the file will be created. If size is omitted, and path points to an existing file with a size greater than zero, the size of that file will be used. If not, an error is reported. Granularity sets the allocation block size. Defaults to the system page size or the filesystem block size, whichever is larger. Advice tells the kernel how varnishd expects to use this mapped region so that the kernel can choose the appropriate read-ahead and caching techniques. Possible values are normal, random and sequential, corresponding to MADV_NORMAL, MADV_RANDOM and MADV_SEQUENTIAL madvise() advice argument, respectively. Defaults to random.
-s <persistent,path,size> — persistent storage. Varnish will store objects in a file in a manner that will secure the survival of most of the objects in the event of a planned or unplanned shutdown of Varnish. > The persistent storage backend has multiple issues with it and will likely be removed from a future version of Varnish. 💀

🔒 Jail 🔒

Varnish jails are a generalization over various platform specific methods to reduce the privileges of varnish processes. They may have specific options. Available jails are:

-j <solaris[,worker=privspec]> — reduce privileges(5) for varnishd and sub-processes to the minimally required set. Only available on platforms which have the setppriv(2) call. The optional worker argument can be used to pass a privilege-specification (see ppriv(1)) by which to extend the effective set of the varnish worker process. While extended privileges may be required by custom vmods, not using the worker option is always more secure. Example to grant basic privileges to the worker process:

-j solaris,worker=basic

-j <linux[,transparent_hugepage=thp_setting][,unixjailoption...]> — default on Linux platforms, it extends the UNIX jail with Linux-specific mechanisms:
- It warns when workdir is not on a tmpfs.
- It tries to keep the process dumpable after dropping privileges.
- It adds control over the transparent hugepage (THP) setting.

thp_setting can take these values:

ignore: Do nothing
enable: Enable THP (see Note below)
disable: Disable THP
try-disable (default): Try to disable, ignore failure (but emit a warning)

Note: Technically, enable is "disable the disable", so it does not necessarily enable THP. The setting names have been chosen to avoid a confusing double negation. 😭

-j <unix[,user=user][,ccgroup=group][,workuser=user]> — default on all other platforms when varnishd is started with an effective uid of 0 ("as root"). With the unix jail mechanism activated, varnish will switch to an alternative user for subprocesses and change the effective uid of the master process whenever possible. The optional user argument specifies which alternative user to use. It defaults to varnish. The optional ccgroup argument specifies a group to add to varnish subprocesses requiring access to a c-compiler. There is no default. The optional workuser argument specifies an alternative user to use for the worker process. It defaults to vcache. The users given for the user and workuser arguments need to have the same primary ("login") group. To set up a system for the default users with a group name varnish, shell commands similar to these may be used:

groupadd varnish
useradd -g varnish -d /nonexistent -s /bin/false \
  -c "Varnish-Cache Daemon User" varnish
useradd -g varnish -d /nonexistent -s /bin/false \
  -c "Varnish-Cache Worker User" vcache

-j none — last resort jail choice: with jail mechanism none, varnish will run all processes with the privileges it was started with. 🙈

🎀 Management Interface 🎀

If the -T option was specified, varnishd will offer a command-line management interface on the specified address and port. The recommended way of connecting to the command-line management interface is through varnishadm(1).

The commands available are documented in varnish-cli(7).

📜 CLI Command File 📜

The -I option makes it possible to run arbitrary management commands when varnishd is launched, before the worker process is started. In particular, this is the way to load configurations, apply labels to them, and make a VCL instance active that uses those labels on startup:

vcl.load panic /etc/varnish_panic.vcl
vcl.load siteA0 /etc/varnish_siteA.vcl
vcl.load siteB0 /etc/varnish_siteB.vcl
vcl.load siteC0 /etc/varnish_siteC.vcl
vcl.label siteA siteA0
vcl.label siteB siteB0
vcl.label siteC siteC0
vcl.load main /etc/varnish_main.vcl
vcl.use main

Every line in the file, including the last line, must be terminated by a newline or carriage return or is otherwise considered truncated, which is a fatal error. 😱

If a command in the file is prefixed with -, failure will not abort the startup.

Note: it is necessary to include an explicit vcl.use command to select which VCL should be the active VCL when relying on CLI Command File to load the configurations at startup.

⚙️ RUN TIME PARAMETERS ⚙️

Runtime parameters can either be set during startup with the -p command line option for varnishd(1) or through the CLI using the param.set or param.reset commands. They can be locked during startup with the -r command line option. 🔐

📏 Run Time Parameter Units 📏

There are different types of parameters that may accept a list of specific values, or optionally take a unit suffix. Slay.

`bool`

A boolean parameter accepts the values on and off.

It will also recognize the following values:

yes and no
true and false
enable and disable

`bytes`

A bytes parameter requires one of the following units suffixes:

b (bytes)
k (kibibytes, 1024 bytes)
m (mebibytes, 1024 kibibytes)
g (gibibytes, 1024 mebibytes)
t (tebibytes, 1024 gibibytes)
p (pebibytes, 1024 tebibytes)

Multiplicator units may be appended with an extra b. For example 32k is equivalent to 32kb. Bytes units are case-insensitive. 💅

`seconds`

A duration parameter may accept the following units suffixes:

ms (milliseconds)
s (seconds)
m (minutes)
h (hours)
d (days)
w (weeks)
y (years)

If the parameter is a timeout or a deadline, a value of "never" (when allowed) disables the effect of the parameter. ⏳

🚩 Run Time Parameter Flags 🚩

Runtime parameters are marked with shorthand flags to avoid repeating the same text over and over in the table below. The meaning of the flags are:

experimental — we have no solid information about good/bad/optimal values for this parameter. Feedback with experience and observations are most welcome. 🫶
delayed — this parameter can be changed on the fly, but will not take effect immediately.
restart — the worker process must be stopped and restarted, before this parameter takes effect.
reload — the VCL programs must be reloaded for this parameter to take effect.
wizard — do not touch unless you really know what you're doing. 🧙
only_root — only works if varnishd is running as root.

🌸 Default Value Exceptions on 32 bit Systems 🌸

Be aware that on 32 bit systems, certain default or maximum values are reduced relative to the values listed below, in order to conserve VM space:

workspace_client: 24k
workspace_backend: 20k
http_resp_size: 8k
http_req_size: 12k
gzip_buffer: 4k
vsl_buffer: 4k
vsl_space: 1G (maximum)
thread_pool_stack: 64k

📋 List of Parameters 📋

This text is produced from the same text you will find in the CLI if you use the param.show command:

⭐ `accept_filter`

NB: This parameter depends on a feature which is not available on all platforms.

Units: bool
Default: on (if your platform supports accept filters)

Enable kernel accept-filters. This may require a kernel module to be loaded to have an effect when enabled.

Okay so quick heads up bestie: turning on accept_filter might straight-up stop some requests from ever reaching Varnish in the first place. Malformed requests can sneak by unnoticed and won't bump the client_req_400 counter. And GET or HEAD requests that come with a body? Yeah, those can get blocked entirely. 💀

acceptor_sleep_decay — the "chill out faster" knob. Default: 0.9, Minimum: 0, Maximum: 1, Flags: experimental.

So if we run out of resources — file descriptors, worker threads, that kind of drama — the acceptor just sleeps between accepts. This parameter multiplicatively shrinks that sleep duration for every successful accept. Like, 0.9 = reduce by 10%. Slay.

acceptor_sleep_incr — Units: seconds, Default: 0.000, Minimum: 0.000, Maximum: 1.000, Flags: experimental.

Same vibe as above: out of resources → acceptor sleeps between accepts. This one controls how much longer we sleep each time we fail to accept a new connection. 😴

acceptor_sleep_max — Units: seconds, Default: 0.050, Minimum: 0.000, Maximum: 10.000, Flags: experimental.

Also out-of-resources acceptor sleeping. This one caps how long it can sleep between attempts to accept new connections. The ceiling, the limit, the periodt. 👑

auto_restart — Units: bool, Default: on.

Automatically restart the child/worker process if it dies. Set it and forget it, bestie.

backend_idle_timeout — Units: seconds, Default: 60.000, Minimum: 1.000.

Timeout before we close unused backend connections. If it's just sitting there unused, it's getting yeeted. 🦋

backend_local_error_holdoff — Units: seconds, Default: 10.000, Minimum: 0.000, Flags: experimental.

When connecting to backends, certain error codes (EADDRNOTAVAIL, EACCESS, EPERM) are signaling a local resource shortage or a config issue. And retrying those connection attempts might just make everything worse, because the kernel operations involved are complex. So this parameter stops repeated connection attempts for however long you configure. Hold up, take a beat.

backend_remote_error_holdoff — Units: seconds, Default: 0.250, Minimum: 0.000, Flags: experimental.

Also when connecting to backends: certain error codes (ECONNREFUSED, ENETUNREACH) signal fundamental connection issues — like the backend not accepting connections, or routing problems — where repeated connection attempts are considered useless. So this parameter prevents repeated connection attempts for the configured duration. No point knocking if nobody's home.

backend_wait_limit — Default: 0, Minimum: 0, Flags: experimental.

Maximum number of transactions that can queue up waiting for a backend connection to become available. The default of 0 (zero) means there is no transaction queueing at all. VCL can override this default value for each backend.

Heads up: this feature must be used with caution, because it can cause threads to pile up and increase response times. Not the vibe. 🙈

backend_wait_timeout — Units: seconds, Default: 0.000, Minimum: 0.000, Flags: timeout, experimental.

When a backend has no connections available for a transaction, that transaction can be queued (see backend_wait_limit) to wait for a connection. This is the default time the transaction will wait before giving up. VCL can override this default value for each backend.

It is strongly advised to never set this higher than a couple of seconds. fr fr.

ban_any_variant — Units: checks, Default: 0, Minimum: 0.

Maximum number of possibly non-matching variants that we evaluate against the ban list during a lookup. Setting this to 0 means only the matching variants get evaluated against the current ban list.

ban_cutoff — Units: bans, Default: 0, Minimum: 0, Flags: experimental.

Expurge long tail content from the cache to keep the number of bans below this value. 0 disables it.

When this parameter is set to a non-zero value, the ban lurker keeps working the ban list as usual, top to bottom — but when it reaches the ban_cutoff-th ban, it treats all objects as if they matched a ban and expunges them from cache. Since actively used objects get tested against the ban list at request time and are thus likely to be associated with bans near the top of the ban list, with ban_cutoff, the least recently accessed objects (the "long tail") get removed.

This parameter is a safety net to avoid bad response times from bans being tested at lookup time. Setting a cutoff trades response time for cache efficiency. The recommended value is proportional to rate(bans_lurker_tests_tested) / n_objects while the ban lurker is working, which is the number of bans the system can sustain. The additional latency due to request ban testing is in the order of ban_cutoff / rate(bans_lurker_tests_tested). For example, for rate(bans_lurker_tests_tested) = 2M/s and a tolerable latency of 100ms, a good value for ban_cutoff may be 200K. Math girlies rise up. ✨

ban_dups — Units: bool, Default: on.

Eliminate older identical bans when a new ban is added. This saves CPU cycles by not comparing objects to identical bans. It's a waste of time if you have many bans which are never identical.

ban_lurker_age — Units: seconds, Default: 60.000, Minimum: 0.000.

The ban lurker will ignore bans until they are this old. When a ban is added, the active traffic gets tested against it as part of object lookup. Because many applications issue bans in bursts, this parameter holds the ban-lurker off until the rush is over. Set it to the approximate time a ban-burst takes. 🎀

ban_lurker_batch — Default: 1000, Minimum: 1.

The ban lurker sleeps ${ban_lurker_sleep} after examining this many objects. Use this to pace the ban-lurker if it's eating too many resources. Snack break, bestie.

ban_lurker_holdoff — Units: seconds, Default: 0.010, Minimum: 0.000, Flags: experimental.

How long the ban lurker sleeps when giving way to lookup due to lock contention.

ban_lurker_sleep — Units: seconds, Default: 0.010, Minimum: 0.000.

How long the ban lurker sleeps after examining ${ban_lurker_batch} objects. Use this to pace the ban-lurker if it's eating too many resources. A value of zero will disable the ban lurker entirely. Gone. Poof. 💀

between_bytes_timeout — Units: seconds, Default: 60.000, Minimum: 0.000, Flags: timeout.

We only wait this many seconds between bytes received from the backend before giving up the fetch. VCL values — per backend or per backend request — take precedence. This parameter does not apply to pipe'ed requests.

cc_command — NB: The actual default value for this parameter depends on the Varnish build environment and options. Default: exec $CC $CFLAGS %w -shared -o %o %s, Flags: must_reload.

The command used for compiling the C source code to a dlopen(3) loadable object. The following expansions can be used:

%s: the source file name
%o: the output file name
%w: the cc_warnings parameter
%d: the raw default cc_command
%D: the expanded default cc_command
%n: the working directory (-n option)
%%: a percent sign

Unknown percent expansion sequences are ignored, and to avoid future incompatibilities, percent characters should be escaped with a double percent sequence.

The %d and %D expansions let you pass the parameter's default value to a wrapper script to perform additional processing. Nested defaults, we love to see it.

cc_warnings — NB: The actual default value for this parameter depends on the Varnish build environment and options. Default: -Wall -Werror, Flags: must_reload.

Warnings used when compiling the C source code with the cc_command parameter. By default, VCL is compiled with the same set of warnings as Varnish itself. Match your energy to the codebase, basically.

cli_limit — Units: bytes, Default: 64k, Minimum: 128b, Maximum: 99999999b.

Maximum size of CLI response. If the response exceeds this limit, the response code will be 201 instead of 200, and the last line will indicate the truncation.

cli_timeout — Units: seconds, Default: 60.000, Minimum: 0.000, Flags: timeout.

Timeout for the child's replies to CLI requests. ⏳

clock_skew — Units: seconds, Default: 10, Minimum: 0.

How much clockskew we are willing to accept between the backend and our own clock.

clock_step — Units: seconds, Default: 1.000, Minimum: 0.000.

How much observed clock step we are willing to accept before we panic. Panic is the threshold here, no cap.

connect_timeout — Units: seconds, Default: 3.500, Minimum: 0.000, Flags: timeout.

Default connection timeout for backend connections. We only try to connect to the backend for this many seconds before giving up. VCL can override this default value for each backend and backend request.

critbit_cooloff — Units: seconds, Default: 180.000, Minimum: 60.000, Maximum: 254.000, Flags: wizard.

How long the critbit hasher keeps deleted objheads on the cooloff list. 🧙 wizard flag, so you know it's serious.

debug — Default: none.

Enable/Disable various kinds of debugging.

none — Disable all debugging

Use +/- prefix to set/reset individual bits:

req_state — VSL Request state engine
workspace — VSL Workspace operations
waitinglist — VSL Waitinglist events
syncvsl — Make VSL synchronous
hashedge — Edge cases in Hash
vclrel — Rapid VCL release
lurker — VSL Ban lurker
esi_chop — Chop ESI fetch to bits
flush_head — Flush after http1 head
vtc_mode — Varnishtest Mode
witness — Emit WITNESS lock records
vsm_keep — Keep the VSM file on restart
slow_acceptor — Slow down Acceptor
h2_nocheck — Disable various H2 checks
vmod_so_keep — Keep copied VMOD libraries
processors — Fetch/Deliver processors
protocol — Protocol debugging
vcl_keep — Keep VCL C and so files
lck — Additional lock statistics

That's a whole menu of little switches, bestie. Flip them with the +/- prefix and go.

default_grace — Units: seconds, Default: 10s, Minimum: 0.000, Flags: obj_sticky.

Default grace period. We will deliver an object this long after it has expired, provided another thread is attempting to get a new copy. Charity for stale objects, lowkey iconic. 💖

`default_keep` 💖

Units: seconds
Default: 0s
Minimum: 0.000
Flags: obj_sticky

Default keep period. Bestie, we will keep a useless object around this long, making it available for conditional backend fetches. That means that the object will be removed from the cache at the end of ttl+grace+keep. 💅

`default_ttl` ✨

Units: seconds
Default: 2m
Minimum: 0.000
Flags: obj_sticky

The TTL assigned to objects if neither the backend nor the VCL code assigns one. Periodt.

`experimental` 🧪

Default: none

Enable/Disable experimental features.

none — Disable all experimental features

Use +/- prefix to set/reset individual bits:

drop_pools — Drop thread pools

`feature` 🌈

Default: none,+https_scheme,+validate_headers,+vcl_req_reset

Enable/Disable various minor features.

default — Set default value (deprecated: use param.reset)
none — Disable all features.

Use +/- prefix to enable/disable individual feature:

http2 — Enable HTTP/2 protocol support.
short_panic — Short panic message.
no_coredump — No coredumps. Must be set before child process starts.
https_scheme — Extract host from full URI in the HTTP/1 request line, if the scheme is https.
http_date_postel — Tolerate non compliant timestamp headers like Date, Last-Modified, Expires etc.
esi_ignore_https — Convert <esi:include src"https://... to http://...
esi_disable_xml_check — Allow ESI processing on non-XML ESI bodies
esi_ignore_other_elements — Ignore XML syntax errors in ESI bodies.
esi_remove_bom — Ignore UTF-8 BOM in ESI bodies.
esi_include_onerror — Parse the onerror attribute of <esi:include> tags.
wait_silo — Wait for persistent silos to completely load before serving requests.
validate_headers — Validate all header set operations to conform to RFC9110 section 5.5.
busy_stats_rate — Make busy workers comply with thread_stats_rate.
trace — Enable VCL tracing by default (enable (be)req.trace). Required for tracing vcl_init / vcl_fini
vcl_req_reset — Stop processing client VCL once the client is gone. When this happens MAIN.req_reset is incremented.

`fetch_chunksize` 📦

Units: bytes
Default: 16k
Minimum: 4k
Flags: experimental

The default chunksize used by fetcher. This should be bigger than the majority of objects with short TTLs. Internal limits in the storage_file module makes increases above 128kb a dubious idea. iykyk

`fetch_maxchunksize` 📦✨

Units: bytes
Default: 0.25G
Minimum: 64k
Flags: experimental

The maximum chunksize we attempt to allocate from storage. Making this too large may cause delays and storage fragmentation. Careful bestie.

`first_byte_timeout` ⏳

Units: seconds
Default: 60.000
Minimum: 0.000
Flags: timeout

Default timeout for receiving first byte from backend. We only wait for this many seconds for the first byte before giving up. VCL can override this default value for each backend and backend request. This parameter does not apply to pipe'ed requests.

`gzip_buffer` 🫧

Units: bytes
Default: 32k
Minimum: 2k
Flags: experimental

Size of malloc buffer used for gzip processing. Size of buffer used for gzip processing. These buffers are used for in-transit data, for instance gunzip'ed data being sent to a client. Making this space to small results in more overhead, writes to sockets etc, making it too big is probably just a waste of memory. Gzip buffers are allocated from Transient storage.

`gzip_level` 🗜️

Default: 6
Minimum: 0
Maximum: 9

Gzip compression level: 0=debug, 1=fast, 9=best

`gzip_memlevel` 🧠

Default: 8
Minimum: 1
Maximum: 9

Gzip memory level 1=slow/least, 9=fast/most compression. Memory impact is 1=1k, 2=2k, ... 9=256k.

`h2_header_table_size` 📋

Units: bytes
Default: 4k
Minimum: 0b

HTTP2 header table size. This is the size that will be used for the HPACK dynamic decoding table.

The value of this parameter defines SETTINGS_HEADER_TABLE_SIZE in the initial SETTINGS frame sent to the client when a new HTTP2 session is established.

`h2_initial_window_size` 🪟

Units: bytes
Default: 65535b
Minimum: 65535b
Maximum: 2147483647b

HTTP2 initial flow control window size.

The value of this parameter defines SETTINGS_INITIAL_WINDOW_SIZE in the initial SETTINGS frame sent to the client when a new HTTP2 session is established.

`h2_max_concurrent_streams` 🌊

Units: streams
Default: 100
Minimum: 0

HTTP2 Maximum number of concurrent streams. This is the number of requests that can be active at the same time for a single HTTP2 connection.

The value of this parameter defines SETTINGS_MAX_CONCURRENT_STREAMS in the initial SETTINGS frame sent to the client when a new HTTP2 session is established.

`h2_max_frame_size` 🧩

Units: bytes
Default: 16k
Minimum: 16k
Maximum: 16777215b

HTTP2 maximum per frame payload size we are willing to accept.

The value of this parameter defines SETTINGS_MAX_FRAME_SIZE in the initial SETTINGS frame sent to the client when a new HTTP2 session is established.

`h2_max_header_list_size` 📏

Units: bytes
Default: 0b
Minimum: 0b
Maximum: 2147483647b

HTTP2 maximum size of an uncompressed header list. This parameter is not mapped to SETTINGS_MAX_HEADER_LIST_SIZE in the initial SETTINGS frame, the http_req_size parameter is instead.

The http_req_size advises HTTP2 clients of the maximum size for the header list. Exceeding http_req_size results in a reset stream after processing the HPACK block to preserve the connection, but exceeding h2_max_header_list_size results in the HTTP2 connection going away immediately.

If h2_max_header_list_size is lower than http_req_size, it has no effect, except for the special value zero interpreted as 150% of http_req_size.

`h2_rapid_reset` 🚨

Units: seconds
Default: 1.000
Minimum: 0.000
Flags: delayed, experimental

The upper threshold for how soon an http/2 RST_STREAM frame has to be parsed after a HEADERS frame for it to be treated as suspect and subjected to the rate limits specified by h2_rapid_reset_limit and h2_rapid_reset_period. Changes to this parameter affect the default for new HTTP2 sessions. vmod_h2(3) can be used to adjust it from VCL.

`h2_rapid_reset_limit` 🚦

Default: 100
Minimum: 0
Flags: delayed, experimental

HTTP2 RST Allowance. Specifies the maximum number of allowed stream resets issued by a client over a time period before the connection is closed. Setting this parameter to 0 disables the limit. Changes to this parameter affect the default for new HTTP2 sessions. vmod_h2(3) can be used to adjust it from VCL.

`h2_rapid_reset_period` ⏱️

Units: seconds
Default: 60.000
Minimum: 1.000
Flags: delayed, experimental, wizard

HTTP2 sliding window duration for h2_rapid_reset_limit. Changes to this parameter affect the default for new HTTP2 sessions. vmod_h2(3) can be used to adjust it from VCL.

`h2_rx_window_increment` 🪄

Units: bytes
Default: 1M
Minimum: 1M
Maximum: 1G
Flags: wizard

HTTP2 Receive Window Increments. How big credits we send in WINDOW_UPDATE frames Only affects incoming request bodies (ie: POST, PUT etc.)

`h2_rx_window_low_water` 🌊

Units: bytes
Default: 10M
Minimum: 65535b
Maximum: 1G
Flags: wizard

HTTP2 Receive Window low water mark. We try to keep the window at least this big Only affects incoming request bodies (ie: POST, PUT etc.)

`h2_rxbuf_storage` 💾

Default: Transient
Flags: must_restart

The name of the storage backend that HTTP/2 receive buffers should be allocated from.

`h2_window_timeout` ⌛

Units: seconds
Default: 5.000
Minimum: 0.000
Flags: timeout, wizard

HTTP2 time limit without window credits. How long a stream may wait for the client to credit the window and allow for more DATA frames to be sent.

`http1_iovs` 🧵

Units: struct iovec (=16 bytes)
Default: 64
Minimum: 5
Maximum: 1024
Flags: wizard

Number of io vectors to allocate for HTTP1 protocol transmission. A HTTP1 header needs 7 + 2 per HTTP header field. Allocated from workspace_thread. This parameter affects only io vectors used for client delivery. For backend fetches, the maximum number of io vectors (up to IOV_MAX) is allocated from available workspace_thread memory.

`http_gzip_support` 🗜️💖

Units: bool
Default: on

Enable gzip support. When enabled Varnish request compressed objects from the backend and store them compressed. If a client does not support gzip encoding Varnish will uncompress compressed objects on demand. Varnish will also rewrite the Accept-Encoding header of clients indicating support for gzip to:

Accept-Encoding: gzip

Clients that do not support gzip will have their Accept-Encoding header removed. For more information on how gzip is implemented please see the chapter on gzip in the Varnish reference.

When gzip support is disabled the variables beresp.do_gzip and beresp.do_gunzip have no effect in VCL.

`http_max_hdr` 📨

Units: header lines
Default: 64
Minimum: 32
Maximum: 65535

Maximum number of HTTP header lines we allow in {req|resp|bereq|beresp}.http (obj.http is autosized to the exact number of headers). Cheap, ~20 bytes, in terms of workspace memory. Note that the first line occupies five header lines.

`http_range_support` 🎯

Units: bool
Default: on

Enable support for HTTP Range headers.

`http_req_hdr_len` 📏

Units: bytes
Default: 8k
Minimum: 40b

Maximum length of any HTTP client request header we will allow. The limit is inclusive its continuation lines.

`http_req_overflow_status` 🚦

Units: HTTP status code or 0 to disable
Default: 0
Minimum: 400
Maximum: 500

okay so this is the HTTP status code that gets sent back if http_req_size gets exceeded. The default value of 0 closes the connection silently without sending a HTTP response at all — just ghosting the client, no cap 💀. And heads up bestie: there is no standard HTTP status which exactly matches the implementation of http_req_size. 414 applies to the URL only, while 413 applies to the request body. 400 is probably the least incorrect alternative value to sending no response at all (0). It can also be seen as an internal error on the Varnish side due to configured limits, so 500 is also allowed.

`http_req_size`

Units: bytes
Default: 32k
Minimum: 0.25k

Maximum number of bytes of HTTP client request we will deal with. This is a limit on all bytes up to the double blank line which ends the HTTP request. The memory for the request is allocated from the client workspace (param: workspace_client) and this parameter limits how much of that the request is allowed to take up. ✨

For HTTP2 clients, it is advertised as MAX_HEADER_LIST_SIZE in the initial SETTINGS frame. iykyk 🦋

`http_resp_hdr_len`

Units: bytes
Default: 8k
Minimum: 40b

Maximum length of any HTTP backend response header we will allow. The limit is inclusive its continuation lines.

`http_resp_size`

Units: bytes
Default: 32k
Minimum: 0.25k

Maximum number of bytes of HTTP backend response we will deal with. This is a limit on all bytes up to the double blank line which ends the HTTP response. The memory for the response is allocated from the backend workspace (param: workspace_backend) and this parameter limits how much of that the response is allowed to take up. 💅

`idle_send_timeout`

Units: seconds
Default: 60.000
Minimum: 0.000
Maximum: 3600.000
Flags: timeout, delayed

Send timeout for individual pieces of data on client connections. May get extended if send_timeout applies.

When this timeout is hit, the session is closed. closed closed. 🚪

See the man page for setsockopt(2) or socket(7) under SO_SNDTIMEO for more information.

`listen_depth`

Units: connections
Default: 1024
Minimum: 0
Flags: must_restart

Listen queue depth.

`lru_interval`

Units: seconds
Default: 2.000
Minimum: 0.000
Flags: experimental

Grace period before object moves on LRU list. Objects are only moved to the front of the LRU list if they have not been moved there already inside this timeout period. This reduces the amount of lock operations necessary for LRU list access. efficiency queen behavior 👑

`max_esi_depth`

Units: levels
Default: 5
Minimum: 0

Maximum depth of esi:include processing.

`max_restarts`

Units: restarts
Default: 4
Minimum: 0
Flags: delayed

Upper limit on how many times a request can restart.

`max_retries`

Units: retries
Default: 4
Minimum: 0
Flags: delayed

Upper limit on how many times a backend fetch can retry.

`max_vcl`

Default: 100
Minimum: 0

Threshold of loaded VCL programs. (VCL labels are not counted.) Parameter max_vcl_handling determines behaviour.

`max_vcl_handling`

Default: 1
Minimum: 0
Maximum: 2

Behaviour when attempting to exceed max_vcl loaded VCL.

0 — Ignore max_vcl parameter.
1 — Issue warning.
2 — Refuse loading VCLs.

`nuke_limit`

Units: allocations
Default: 50
Minimum: 0
Flags: experimental

Maximum number of objects we attempt to nuke in order to make space for a object body. 💥

`panic_buffer`

Units: bytes
Default: 64k
Minimum: 4k
Maximum: 10M
Flags: must_restart

Size of the panic message buffer. The panic buffer is allocated in the working directory as memory shared between the management and worker process, so sufficient working directory space should be accounted for if this value is adjusted. Panic messages are truncated to the configured size. budget accordingly bestie 🙈

`pcre2_depth_limit`

Default: 20
Minimum: 1

The recursion depth-limit for the internal match logic in a pcre2_match().

(See: pcre2_set_depth_limit() in pcre2 docs.)

This puts an upper limit on the amount of stack used by PCRE2 for certain classes of regular expressions.

We have set the default value low in order to prevent crashes, at the cost of possible regexp matching failures. trade-offs, we love them 😭

Matching failures will show up in the log as VCL_Error messages.

`pcre2_jit_compilation`

Units: bool
Default: on

Use the pcre2 JIT compiler if available. 🔥

`pcre2_match_limit`

Default: 10000
Minimum: 1

The limit for the number of calls to the internal match logic in pcre2_match().

(See: pcre2_set_match_limit() in pcre2 docs.)

This parameter limits how much CPU time regular expression matching can soak up.

`ping_interval`

Units: seconds
Default: 3
Minimum: 0
Flags: must_restart

Interval between pings from parent to child. Zero will disable pinging entirely, which makes it possible to attach a debugger to the child. 🐛

`pipe_sess_max`

Units: connections
Default: 0
Minimum: 0

Maximum number of sessions dedicated to pipe transactions.

`pipe_task_deadline`

Units: seconds
Default: 0.000
Minimum: 0.000
Flags: timeout

Deadline for PIPE sessions. Regardless of activity in either direction after this many seconds, the session is closed.

`pipe_timeout`

Units: seconds
Default: 60.000
Minimum: 0.000
Flags: timeout

Idle timeout for PIPE sessions. If nothing have been received in either direction for this many seconds, the session is closed. 💤

`pool_req`

Default: 10,100,10

Parameters for per worker pool request memory pool.

The three numbers are:

min_pool — minimum size of free pool.
max_pool — maximum size of free pool.
max_age — max age of free element.

`pool_sess`

Default: 10,100,10

Parameters for per worker pool session memory pool.

The three numbers are:

min_pool — minimum size of free pool.
max_pool — maximum size of free pool.
max_age — max age of free element.

`pool_vbo`

Default: 10,100,10

Parameters for backend object fetch memory pool.

The three numbers are:

min_pool — minimum size of free pool.
max_pool — maximum size of free pool.
max_age — max age of free element.

`prefer_ipv6`

Units: bool
Default: off

Prefer IPv6 address when connecting to backends which have both IPv4 and IPv6 addresses. 🌈

`rush_exponent`

Units: requests per request
Default: 3
Minimum: 1
Flags: experimental

How many parked request we start for each completed request on the object. NB: Even with the implicit delay of delivery, this parameter controls an exponential increase in number of worker threads. A value of 1 will instead serialize requests resumption and is only useful for testing purposes. exponential, girlies. be normal about this one. 📈

`send_timeout`

Units: seconds
Default: 600.000
Minimum: 0.000
Flags: timeout, delayed

Total timeout for ordinary HTTP1 responses. Does not apply to some internally generated errors and pipe mode.

When idle_send_timeout is hit while sending an HTTP1 response, the timeout is extended unless the total time already taken for sending the response in its entirety exceeds this many seconds.

When this timeout is hit, the session is closed 💀

`shortlived`

Units: seconds
Default: 10.000
Minimum: 0.000

Objects created with (ttl+grace+keep) shorter than this are always put in transient storage.

`sigsegv_handler`

Units: bool
Default: on
Flags: must_restart

Install a signal handler which tries to dump debug information on segmentation faults, bus errors and abort signals. 🚨

`startup_timeout`

Units: seconds
Default: 0.000
Minimum: 0.000
Flags: timeout

Alternative timeout for the initial worker process startup. If cli_timeout is longer than startup_timeout, it is used instead.

`syslog_cli_traffic`

Units: bool
Default: on

Log all CLI traffic to syslog(LOG_INFO).

`tcp_fastopen`

NB: This parameter depends on a feature which is not available on all platforms. 🫶

Units: bool
Default: off
Flags: must_restart

Enable TCP Fast Open extension.

`tcp_keepalive_intvl`

NB: This parameter depends on a feature which is not available on all platforms. 🫶

Units: seconds
Default: platform dependent
Minimum: 1.000
Maximum: 100.000
Flags: experimental

The number of seconds between TCP keep-alive probes. Ignored for Unix domain sockets.

`tcp_keepalive_probes`

NB: This parameter depends on a feature which is not available on all platforms. 🫶

Units: probes
Default: platform dependent
Minimum: 1
Maximum: 100
Flags: experimental

The maximum number of TCP keep-alive probes to send before giving up and killing the connection if no response is obtained from the other end. Ignored for Unix domain sockets.

`tcp_keepalive_time`

NB: This parameter depends on a feature which is not available on all platforms. 🫶

Units: seconds
Default: platform dependent
Minimum: 1.000
Maximum: 7200.000
Flags: experimental

The number of seconds a connection needs to be idle before TCP begins sending out keep-alive probes. Ignored for Unix domain sockets. 👀

`thread_pool_add_delay`

Units: seconds
Default: 0.000
Minimum: 0.000
Flags: experimental

Wait at least this long after creating a thread.

Some (buggy) systems may need a short (sub-second) delay between creating threads. Set this to a few milliseconds if you see the threads_failed counter grow too much.

Setting this too high results in insufficient worker threads. ⚠️

`thread_pool_destroy_delay`

Units: seconds
Default: 1.000
Minimum: 0.010
Flags: delayed, experimental

Wait this long after destroying a thread.

This controls the decay of thread pools when idle(-ish). 🌙

thread_pool_fail_delay
- Units: seconds
- Default: 0.200
- Minimum: 0.010
- Flags: experimental

Okay so when a thread creation flops, we wait at at least this long before we even try to make another one. Like, chill, breathe, we'll try again in a sec 💅

And why tho? Because failing to create a worker thread is lowkey a sign that the end is near — the process is running out of some resource. So this delay is literally just us not rushing the apocalypse unnecessarily. Main character restraint.

If thread creation failures are being a problem, check that thread_pool_max isn't too high.

It may also help to increase thread_pool_timeout and thread_pool_min, to reduce the rate at which treads are destroyed and later recreated.

thread_pool_max
- Units: threads
- Default: 5000
- Minimum: thread_pool_min
- Flags: delayed

The maximum number of worker threads in each pool.

Do not set this higher than you have to, bestie. Excess worker threads soak up RAM and CPU and generally just get in the way of getting work done. It's giving clutter 🫶

thread_pool_min
- Units: threads
- Default: 100
- Minimum: 5
- Maximum: thread_pool_max
- Flags: delayed

The minimum number of worker threads in each pool.

Bumping this up may help you ramp up faster from low load situations, or when threads have expired.

Technical minimum is 5 threads, but this parameter is strongly recommended to be at least 10.

thread_pool_reserve
- Units: threads
- Default: 0 (auto-tune: 5% of thread_pool_min)
- Maximum: 95% of thread_pool_min
- Flags: delayed

The number of worker threads reserved for vital tasks in each pool.

So here's the tea: tasks may require other tasks to complete (for example, client requests may require backend requests, http2 sessions require streams, which require requests). This reserve is to ensure that lower priority tasks do not prevent higher priority tasks from running even under high load. It's literally the VIP section ✨

The effective value is at least 5 (the number of internal priority classes), irrespective of this parameter.

thread_pool_stack
- Units: bytes
- Default: 80k
- Minimum: sysconf(_SC_THREAD_STACK_MIN)
- Flags: delayed

Worker thread stack size. This will likely be rounded up to a multiple of 4k (or whatever the page_size might be) by the kernel. Not us, the kernel 💀

The required stack size is primarily driven by the depth of the call-tree. The most common relevant determining factors in varnish core code are GZIP (un)compression, ESI processing and regular expression matches. VMODs may also require significant amounts of additional stack. The nesting depth of VCL subs is another factor, although typically not predominant.

The stack size is per thread, so the maximum total memory required for worker thread stacks is in the order of size = thread_pools x thread_pool_max x thread_pool_stack.

Thus, in particular for setups with many threads, keeping the stack size at a minimum helps reduce the amount of memory required by Varnish.

On the other hand — and this part matters fr — thread_pool_stack must be large enough under all circumstances, otherwise varnish will crash due to a stack overflow. Usually, a stack overflow manifests itself as a segmentation fault (aka segfault / SIGSEGV) with the faulting address being near the stack pointer (sp).

Unless stack usage can be reduced, thread_pool_stack must be increased when a stack overflow occurs. Setting it in 150%-200% increments is recommended until stack overflows cease to occur.

thread_pool_timeout
- Units: seconds
- Default: 300.000
- Minimum: 10.000
- Flags: delayed, experimental

Thread idle threshold.

Threads in excess of thread_pool_min, which have been idle for at least this long, will be destroyed. Sorry bestie, if you're not pulling your weight, you're out 🙈

thread_pool_watchdog
- Units: seconds
- Default: 60.000
- Minimum: 0.100
- Flags: experimental

Thread queue stuck watchdog.

If no queued work have been released for this long, the worker process panics itself. Yep. It's the self-destruct button, but for a jammed queue 💀

thread_pools
- Units: pools
- Default: 2
- Minimum: 1
- Maximum: 32
- Flags: delayed, experimental

Number of worker thread pools.

Increasing the number of worker pools decreases lock contention. Each worker pool also has a thread accepting new connections, so for very high rates of incoming new connections on systems with many cores, increasing the worker pools may be required.

Too many pools waste CPU and RAM resources, and more than one pool for each CPU is most likely detrimental to performance. Moderation, queen 👑

Can be increased on the fly, but decreases require a restart to take effect, unless the drop_pools experimental debug flag is set.

thread_queue_limit
- Units: requests
- Default: 20
- Minimum: 0
- Flags: experimental

Permitted request queue length per thread-pool.

This sets the number of requests we will queue, waiting for an available thread. Above this limit sessions will be dropped instead of queued. No queue spot? No entry 💅

thread_stats_rate
- Units: requests
- Default: 10
- Minimum: 0
- Flags: experimental

Worker threads accumulate statistics, and dump these into the global stats counters if the lock is free when they finish a job (request/fetch etc.) This parameters defines the maximum number of jobs a worker thread may handle, before it is forced to dump its accumulated stats into the global counters. Like a lil journal entry, but for stats 📓

timeout_idle
- Units: seconds
- Default: 5.000
- Minimum: 0.000
- Maximum: 3600.000

Idle timeout for client connections.

A connection is considered idle until we have received the full request headers.

This parameter is particularly relevant for HTTP1 keepalive connections which are closed unless the next request is received before this timeout is reached.

timeout_linger
- Units: seconds
- Default: 0.050
- Minimum: 0.000
- Flags: experimental

How long the worker thread lingers on an idle session before handing it over to the waiter. When sessions are reused, as much as half of all reuses happen within the first 100 msec of the previous request completing. Setting this too high results in worker threads not doing anything for their keep, setting it too low just means that more sessions take a detour around the waiter. It's giving "should I stay or should I go" 😭

transit_buffer
- Units: bytes
- Default: 0b
- Minimum: 0b

The number of bytes which Varnish buffers for uncacheable backend streaming fetches — in other words, how many bytes Varnish reads from the backend ahead of what has been sent to the client. A zero value means no limit, the object is fetched as fast as possible.

When dealing with slow clients, setting this parameter to non-zero can prevent large uncacheable objects from being stored in full when the intent is to simply stream them to the client. As a result, a slow client transaction holds onto a backend connection until the end of the delivery.

This parameter is the default to the VCL variable beresp.transit_buffer, which can be used to control the transit buffer per backend request.

uncacheable_ttl
- Units: seconds
- Default: 2m
- Minimum: 0.000
- Flags: obj_sticky

The TTL assigned to uncacheable objects by the built-in VCL.

vary_notice
- Units: variants
- Default: 10
- Minimum: 1

How many variants need to be evaluated to log a Notice that there might be too many variants. Basically the "hey bestie, this is getting out of hand" alarm 🌸

vcc_allow_inline_c — Deprecated alias for the vcc_feature parameter.
vcc_err_unref — Deprecated alias for the vcc_feature parameter.
vcc_feature
- Default: none,+err_unref,+unsafe_path

Enable/Disable various VCC behaviors.

default — Set default value (deprecated: use param.reset)
none — Disable all behaviors.

Use +/- prefix to enable/disable individual behavior:

err_unref — Unreferenced VCL objects result in error.
allow_inline_c — Allow inline C code in VCL.
unsafe_path — Allow / in vmod & include paths. Allow import ... from ....

vcc_unsafe_path — Deprecated alias for the vcc_feature parameter.
vcl_cooldown
- Units: seconds
- Default: 600.000
- Minimum: 1.000

How long a VCL is kept warm after being replaced as the active VCL (granularity approximately 30 seconds). Like letting the old one nap on the couch before it has to move out 🛋️

vcl_path

NB: The actual default value for this parameter depends on the Varnish build environment and options.

Default: ${sysconfdir}/varnish:${datadir}/varnish/vcl

Directory (or colon separated list of directories) from which relative VCL filenames (vcl.load and include) are to be found. By default Varnish searches VCL files in both the system configuration and shared data directories to allow packages to drop their VCL files in a standard location where relative includes would work. Includes using +glob cannot be searched in vcl_path.

vmod_path

NB: The actual default value for this parameter depends on the Varnish build environment and options.

Default: ${libdir}/varnish/vmods

Directory (or colon separated list of directories) where VMODs are to be found.

vsl_buffer
- Units: bytes
- Default: 16k
- Minimum: vsl_reclen + 12 bytes

Bytes of (req-/backend-)workspace dedicated to buffering VSL records. When this parameter is adjusted, most likely workspace_client and workspace_backend will have to be adjusted by the same amount.

Setting this too high costs memory, setting it too low will cause more VSL flushes and likely increase lock-contention on the VSL mutex. Balance, bestie ⚖️

vsl_mask
- Default:

all,-Debug,-ObjProtocol,-ObjStatus,-ObjReason,-ObjHeader,-ExpKill,-WorkThread,-Hash,-VfpAcct,-H2RxHdr,-H2RxBody,-H2TxHdr,-H2TxBody,-VdpAcct,-ReqTarget

Mask individual VSL messages from being logged.

all — Enable all tags
default — Set default value (deprecated: use param.reset)

Use +/- prefix in front of VSL tag name to unmask/mask individual VSL messages. See vsl(7) for possible values.

vsl_reclen
- Units: bytes
- Default: 255b
- Minimum: 16b
- Maximum: vsl_buffer - 12 bytes

Maximum number of bytes in SHM log record.

vsl_space
- Units: bytes
- Default: 80M
- Minimum: 1M
- Maximum: 4G
- Flags: must_restart

The amount of space to allocate for the VSL fifo buffer in the VSM memory segment. If you make this too small, varnish{ncsa|log} etc will not be able to keep up. Making it too large just costs memory resources.

vsm_free_cooldown
- Units: seconds
- Default: 60.000
- Minimum: 10.000

…okay so finishing off the parameter we were <em>literally</em> just on:

Maximum: 600.000

That's how long VSM memory is kept warm after a deallocation 💖 — and heads up, the granularity is approximately 2 seconds, so it's giving "vibes-based precision," not atomic-clock precision. 🫶

`workspace_backend`

Units: bytes
Default: 96k
Minimum: 1k
Flags: delayed

Bytes of HTTP protocol workspace for backend HTTP req/resp. ✨ If it's larger than 4k, use a multiple of 4k for VM efficiency — trust, your VM will thank you. 💅

`workspace_client`

Units: bytes
Default: 96k
Minimum: 9k
Flags: delayed

Bytes of HTTP protocol workspace for clients HTTP req/resp. Use a multiple of 4k for VM efficiency. 🦋

For HTTP/2 compliance this <strong>must</strong> be at least 20k, so you can receive fullsize (=16k) frames from the client. That usually happens only in POST/PUT bodies tho — for other traffic-patterns smaller values work just fine. no cap. 🎀

`workspace_session`

Units: bytes
Default: 0.75k
Minimum: 384b
Flags: delayed

Allocation size for session structure and workspace. The workspace is primarily used for TCP connection addresses. 🌸 If it's larger than 4k, use a multiple of 4k for VM efficiency (yes bestie, same rule, it's iconic for a reason).

`workspace_thread`

Units: bytes
Default: 2k
Minimum: 0.25k
Maximum: 8k
Flags: delayed

Bytes of auxiliary workspace per thread. This workspace is used for certain temporary data structures during the operation of a worker thread. One use is for the IO-vectors used during delivery. 👑

Setting this parameter too low may increase the number of writev() syscalls, and setting it too high just wastes space. 😭 ~0.1k + UIO_MAXIOV * sizeof(struct iovec) (typically = ~16k for 64bit) is considered the maximum sensible value under any known circumstances (excluding exotic vmod use). periodt. 💜

💀 EXIT CODES 💀

Varnish and bundled tools will, in most cases, exit with one of the following codes:

0 — OK ✅
1 — Some error which could be system-dependent and/or transient 🫠
2 — Serious configuration / parameter error — retrying with the same configuration / parameters is most likely useless 🙈

The varnishd master process may also OR its exit code:

with 0x20 when the varnishd child process died 💀
with 0x40 when the varnishd child process was terminated by a signal ⚡
with 0x80 when a core was dumped 🕳️

🌈 SEE ALSO 🌈

🦋 HISTORY 🦋

The varnishd daemon was developed by Poul-Henning Kamp in cooperation with Verdens Gang AS and Varnish Software.

This manual page was written by Dag-Erling Smørgrav with updates by Stig Sandbeck Mathisen < <ssm@debian.org> >, Nils Goroll and others. 💖

👑 COPYRIGHT 👑

This document is licensed under the same licence as Varnish itself. See LICENCE for details. 📜

and that's the whole page bestie!! 🎀 the footer said <code>VARNISHD(1)</code> — it's from the Varnish source, version-stamped and dated like a little time capsule, sitting right at the bottom where footers live. you made it to the end, main character behaviour fr fr, go drink some water and slay 💅✨🫶

📦 where this page comes from

fedora 44: varnish 8.0.2-1.fc44 · BSD-2-Clause AND (BSD-2-Clause-FreeBSD AND BSD-3-Clause AND LicenseRef-Fedora-Public-Domain AND Zlib)
fedora 44: varnish 8.0.2-1.fc44 · BSD-2-Clause AND (BSD-2-Clause-FreeBSD AND BSD-3-Clause AND LicenseRef-Fedora-Public-Domain AND Zlib)
📜 the page's license: BSD-2-Clause
💖 girlified by deepseek-ai/DeepSeek-V4.1-Flash on 2026-10-18: a language model's rewrite of the original, under its terms

VARNISHD(1)

👑 VARNISHD 👑 (that's the manual page VARNISHD(1), General Commands Manual, section 1, iykyk)

🌸 NAME 🌸

✨ SYNOPSIS ✨

🦋 DESCRIPTION 🦋

💖 OPTIONS 💖

🎀 Basic options 🎀

📚 Documentation options 📚

⚙️ Operations options ⚙️

🎚️ Tuning options 🎚️

🔒 Security options 🔒

🧪 Advanced, development and debugging options 🧪

#️⃣ Hash Algorithm #️⃣

💾 Storage Backend 💾

🔒 Jail 🔒

🎀 Management Interface 🎀

📜 CLI Command File 📜

⚙️ RUN TIME PARAMETERS ⚙️

📏 Run Time Parameter Units 📏

bool

bytes

seconds

🚩 Run Time Parameter Flags 🚩

🌸 Default Value Exceptions on 32 bit Systems 🌸

📋 List of Parameters 📋

⭐ accept_filter

default_keep 💖

default_ttl ✨

experimental 🧪

feature 🌈

fetch_chunksize 📦

fetch_maxchunksize 📦✨

first_byte_timeout ⏳

gzip_buffer 🫧

gzip_level 🗜️

gzip_memlevel 🧠

h2_header_table_size 📋

h2_initial_window_size 🪟

h2_max_concurrent_streams 🌊

h2_max_frame_size 🧩

h2_max_header_list_size 📏

h2_rapid_reset 🚨

h2_rapid_reset_limit 🚦

h2_rapid_reset_period ⏱️

h2_rx_window_increment 🪄

h2_rx_window_low_water 🌊

h2_rxbuf_storage 💾

h2_window_timeout ⌛

http1_iovs 🧵

http_gzip_support 🗜️💖

http_max_hdr 📨

http_range_support 🎯

http_req_hdr_len 📏

http_req_overflow_status 🚦

http_req_size

http_resp_hdr_len

http_resp_size

idle_send_timeout

listen_depth

lru_interval

max_esi_depth

max_restarts

max_retries

max_vcl

max_vcl_handling

nuke_limit

panic_buffer

pcre2_depth_limit

pcre2_jit_compilation

pcre2_match_limit

ping_interval

pipe_sess_max

pipe_task_deadline

pipe_timeout

pool_req

pool_sess

pool_vbo

prefer_ipv6

rush_exponent

send_timeout

👑 VARNISHD 👑 (that's the manual page `VARNISHD(1)`, General Commands Manual, section 1, iykyk)

`bool`

`bytes`

`seconds`

⭐ `accept_filter`

`default_keep` 💖

`default_ttl` ✨

`experimental` 🧪

`feature` 🌈

`fetch_chunksize` 📦

`fetch_maxchunksize` 📦✨

`first_byte_timeout` ⏳

`gzip_buffer` 🫧

`gzip_level` 🗜️

`gzip_memlevel` 🧠

`h2_header_table_size` 📋

`h2_initial_window_size` 🪟

`h2_max_concurrent_streams` 🌊

`h2_max_frame_size` 🧩

`h2_max_header_list_size` 📏

`h2_rapid_reset` 🚨

`h2_rapid_reset_limit` 🚦

`h2_rapid_reset_period` ⏱️

`h2_rx_window_increment` 🪄

`h2_rx_window_low_water` 🌊

`h2_rxbuf_storage` 💾

`h2_window_timeout` ⌛

`http1_iovs` 🧵

`http_gzip_support` 🗜️💖

`http_max_hdr` 📨

`http_range_support` 🎯

`http_req_hdr_len` 📏

`http_req_overflow_status` 🚦

`http_req_size`

`http_resp_hdr_len`

`http_resp_size`

`idle_send_timeout`

`listen_depth`

`lru_interval`

`max_esi_depth`

`max_restarts`

`max_retries`

`max_vcl`

`max_vcl_handling`

`nuke_limit`

`panic_buffer`

`pcre2_depth_limit`

`pcre2_jit_compilation`

`pcre2_match_limit`

`ping_interval`

`pipe_sess_max`

`pipe_task_deadline`

`pipe_timeout`

`pool_req`

`pool_sess`

`pool_vbo`

`prefer_ipv6`

`rush_exponent`

`send_timeout`

`shortlived`

`sigsegv_handler`

`startup_timeout`

`syslog_cli_traffic`

`tcp_fastopen`

`tcp_keepalive_intvl`

`tcp_keepalive_probes`

`tcp_keepalive_time`

`thread_pool_add_delay`

`thread_pool_destroy_delay`

`workspace_backend`

`workspace_client`

`workspace_session`

`workspace_thread`