1VARNISHD(1) VARNISHD(1)
2
6 varnishd - HTTP accelerator daemon
7
9 varnishd [-a
10 [name=][address][:port][,PROTO][,user=<user>][,group=<group>][,mode=<mode>]]
11 [-b [host[:port]|path]] [-C] [-d] [-F] [-f config] [-h type[,options]]
12 [-I clifile] [-i identity] [-j jail[,jailoptions]] [-l vsl] [-M
13 address:port] [-n name] [-P file] [-p param=value] [-r
14 param[,param...]] [-S secret-file] [-s [name=]kind[,options]] [-T
15 address[:port]] [-t TTL] [-V] [-W waiter]
16 varnishd [-x parameter|vsl|cli|builtin|optstring]
18 varnishd [-?]
20
22 The varnishd daemon accepts HTTP requests from clients, passes them on
23 to a backend server and caches the returned documents to better satisfy
24 future requests for the same document.
25
27 Basic options
28 -a
29 <[name=][address][:port][,PROTO][,user=<user>][,group=<group>][,mode=<mode>]>
30 Listen for client requests on the specified address and port. The
31 address can be a host name ("localhost"), an IPv4 dotted-quad
32 ("127.0.0.1"), an IPv6 address enclosed in square brackets
33 ("[::1]"), or a path beginning with a '/' for a Unix domain socket
34 ("/path/to/listen.sock"). If address is not specified, varnishd will
35 listen on all available IPv4 and IPv6 interfaces. If port is not
36 specified, port 80 (http) is used. At least one of address or port
37 is required.
38 If a Unix domain socket is specified as the listen address, then the
40 user, group and mode sub-arguments may be used to specify the per‐
41 missions of the socket file -- use names for user and group, and a
42 3-digit octal value for mode. These sub-arguments are not permitted
43 if an IP address is specified. When Unix domain socket listeners are
44 in use, all VCL configurations must have version >= 4.1.
45 Name is referenced in logs. If name is not specified, "a0", "a1",
47 etc. is used. An additional protocol type can be set for the listen‐
48 ing socket with PROTO. Valid protocol types are: HTTP/1 (default),
49 and PROXY.
50 Multiple listening addresses can be specified by using different -a
52 arguments.
53 -b <[host[:port]|path]>
55 Use the specified host as backend server. If port is not speci‐
56 fied, the default is 8080.
57 If the value of -b begins with /, it is interpreted as the abso‐
59 lute path of a Unix domain socket to which Varnish connects. In
60 that case, the value of -b must satisfy the conditions required
61 for the .path field of a backend declaration, see vcl(7). Back‐
62 ends with Unix socket addresses may only be used with VCL ver‐
63 sions >= 4.1.
64 -b can be used only once, and not together with f.
66 -f config
68 Use the specified VCL configuration file instead of the builtin
69 default. See vcl(7) for details on VCL syntax.
70 If a single -f option is used, then the VCL instance loaded from
72 the file is named "boot" and immediately becomes active. If more
73 than one -f option is used, the VCL instances are named "boot0",
74 "boot1" and so forth, in the order corresponding to the -f argu‐
75 ments, and the last one is named "boot", which becomes active.
76 Either -b or one or more -f options must be specified, but not
78 both, and they cannot both be left out, unless -d is used to
79 start varnishd in debugging mode. If the empty string is speci‐
80 fied as the sole -f option, then varnishd starts without start‐
81 ing the worker process, and the management process will accept
82 CLI commands. You can also combine an empty -f option with an
83 initialization script (-I option) and the child process will be
84 started if there is an active VCL at the end of the initializa‐
85 tion.
86 When used with a relative file name, config is searched in the
88 vcl_path. It is possible to set this path prior to using -f
89 options with a -p option. During startup, varnishd doesn't com‐
90 plain about unsafe VCL paths: unlike the varnish-cli(7) that
91 could later be accessed remotely, starting varnishd requires
92 local privileges.
93 -n name
95 Specify the name for this instance. This name is used to con‐
96 struct the name of the directory in which varnishd keeps tempo‐
97 rary files and persistent state. If the specified name begins
98 with a forward slash, it is interpreted as the absolute path to
99 the directory.
100 Documentation options
102 For these options, varnishd prints information to standard output and
103 exits. When a -x option is used, it must be the only option (it outputs
104 documentation in reStructuredText, aka RST).
105 -?
107 Print the usage message.
108 -x parameter
110 Print documentation of the runtime parameters (-p options), see
111 List of Parameters.
112 -x vsl Print documentation of the tags used in the Varnish shared mem‐
114 ory log, see vsl(7).
115 -x cli Print documentation of the command line interface, see var‐
117 nish-cli(7).
118 -x builtin
120 Print the contents of the default VCL program builtin.vcl.
121 -x optstring
123 Print the optstring parameter to getopt(3) to help writing wrap‐
124 per scripts.
125 Operations options
127 -F Do not fork, run in the foreground. Only one of -F or -d can be
128 specified, and -F cannot be used together with -C.
129 -T <address[:port]>
131 Offer a management interface on the specified address and port.
132 See varnish-cli(7) for documentation of the management commands.
133 To disable the management interface use none.
134 -M <address:port>
136 Connect to this port and offer the command line interface.
137 Think of it as a reverse shell. When running with -M and there
138 is no backend defined the child process (the cache) will not
139 start initially.
140 -P file
142 Write the PID of the process to the specified file.
143 -i identity
145 Specify the identity of the Varnish server. This can be accessed
146 using server.identity from VCL and with VSM_Name() from utili‐
147 ties. If not specified the output of gethostname(3) is used.
148 -I clifile
150 Execute the management commands in the file given as clifile
151 before the the worker process starts, see CLI Command File.
152 Tuning options
154 -t TTL Specifies the default time to live (TTL) for cached objects.
155 This is a shortcut for specifying the default_ttl run-time
156 parameter.
157 -p <param=value>
159 Set the parameter specified by param to the specified value, see
160 List of Parameters for details. This option can be used multiple
161 times to specify multiple parameters.
162 -s <[name=]type[,options]>
164 Use the specified storage backend. See Storage Backend section.
165 This option can be used multiple times to specify multiple stor‐
167 age files. Name is referenced in logs, VCL, statistics, etc. If
168 name is not specified, "s0", "s1" and so forth is used.
169 -l <vsl>
171 Specifies size of the space for the VSL records. Scaling suf‐
172 fixes like 'K' and 'M' can be used up to (G)igabytes. Default
173 is 80 Megabytes.
174 Security options
176 -r <param[,param...]>
177 Make the listed parameters read only. This gives the system
178 administrator a way to limit what the Varnish CLI can do. Con‐
179 sider making parameters such as cc_command, vcc_allow_inline_c
180 and vmod_path read only as these can potentially be used to
181 escalate privileges from the CLI.
182 -S secret-file
184 Path to a file containing a secret used for authorizing access
185 to the management port. To disable authentication use none.
186 If this argument is not provided, a secret drawn from the system
188 PRNG will be written to a file called _.secret in the working
189 directory (see opt_n) with default ownership and permissions of
190 the user having started varnish.
191 Thus, users wishing to delegate control over varnish will proba‐
193 bly want to create a custom secret file with appropriate permis‐
194 sions (ie. readable by a unix group to delegate control to).
195 -j <jail[,jailoptions]>
197 Specify the jailing mechanism to use. See Jail section.
198 Advanced, development and debugging options
200 -d Enables debugging mode: The parent process runs in the fore‐
201 ground with a CLI connection on stdin/stdout, and the child
202 process must be started explicitly with a CLI command. Terminat‐
203 ing the parent process will also terminate the child.
204 Only one of -d or -F can be specified, and -d cannot be used
206 together with -C.
207 -C Print VCL code compiled to C language and exit. Specify the VCL
209 file to compile with the -f option. Either -f or -b must be used
210 with -C, and -C cannot be used with -F or -d.
211 -V Display the version number and exit. This must be the only
213 option.
214 -h <type[,options]>
216 Specifies the hash algorithm. See Hash Algorithm section for a
217 list of supported algorithms.
218 -W waiter
220 Specifies the waiter type to use.
221 Hash Algorithm
223 The following hash algorithms are available:
224 -h critbit
226 self-scaling tree structure. The default hash algorithm in Var‐
227 nish Cache 2.1 and onwards. In comparison to a more traditional
228 B tree the critbit tree is almost completely lockless. Do not
229 change this unless you are certain what you're doing.
230 -h simple_list
232 A simple doubly-linked list. Not recommended for production
233 use.
234 -h <classic[,buckets]>
236 A standard hash table. The hash key is the CRC32 of the object's
237 URL modulo the size of the hash table. Each table entry points
238 to a list of elements which share the same hash key. The buckets
239 parameter specifies the number of entries in the hash table.
240 The default is 16383.
241 Storage Backend
243 The following storage types are available:
244 -s <default[,size]>
246 The default storage type resolves to umem where available and
247 malloc otherwise.
248 -s <malloc[,size]>
250 malloc is a memory based backend.
251 -s <umem[,size]>
253 umem is a storage backend which is more efficient than malloc on
254 platforms where it is available.
255 See the section on umem in chapter Storage backends of The Var‐
257 nish Users Guide for details.
258 -s <file,path[,size[,granularity[,advice]]]>
260 The file backend stores data in a file on disk. The file will be
261 accessed using mmap. Note that this storage provide no cache
262 persistence.
263 The path is mandatory. If path points to a directory, a tempo‐
265 rary file will be created in that directory and immediately
266 unlinked. If path points to a non-existing file, the file will
267 be created.
268 If size is omitted, and path points to an existing file with a
270 size greater than zero, the size of that file will be used. If
271 not, an error is reported.
272 Granularity sets the allocation block size. Defaults to the sys‐
274 tem page size or the filesystem block size, whichever is larger.
275 Advice tells the kernel how varnishd expects to use this mapped
277 region so that the kernel can choose the appropriate read-ahead
278 and caching techniques. Possible values are normal, random and
279 sequential, corresponding to MADV_NORMAL, MADV_RANDOM and
280 MADV_SEQUENTIAL madvise() advice argument, respectively.
281 Defaults to random.
282 -s <persistent,path,size>
284 Persistent storage. Varnish will store objects in a file in a
285 manner that will secure the survival of most of the objects in
286 the event of a planned or unplanned shutdown of Varnish. The
287 persistent storage backend has multiple issues with it and will
288 likely be removed from a future version of Varnish.
289 You can also prefix the type with NAME= to explicitly name a storage:
291 -s myStorage=malloc,5G
293 This allows to address it more easily in VCL:
295 set beresp.storage = storage.myStorage;
297 If the name is omitted, Varnish will name storages sN, starting with s0
299 and incrementing N for every new storage.
300 Jail
302 Varnish jails are a generalization over various platform specific meth‐
303 ods to reduce the privileges of varnish processes. They may have spe‐
304 cific options. Available jails are:
305 -j solaris
307 Reduce privileges(5) for varnishd and sub-process to the mini‐
308 mally required set. Only available on platforms which have the
309 setppriv(2) call.
310 -j <unix[,user=`user`][,ccgroup=`group`][,workuser=`user`]>
312 Default on all other platforms when varnishd is started with an
313 effective uid of 0 ("as root").
314 With the unix jail mechanism activated, varnish will switch to
316 an alternative user for subprocesses and change the effective
317 uid of the master process whenever possible.
318 The optional user argument specifies which alternative user to
320 use. It defaults to varnish.
321 The optional ccgroup argument specifies a group to add to var‐
323 nish subprocesses requiring access to a c-compiler. There is no
324 default.
325 The optional workuser argument specifies an alternative user to
327 use for the worker process. It defaults to vcache.
328 -j none
330 last resort jail choice: With jail mechanism none, varnish will
331 run all processes with the privileges it was started with.
332 Management Interface
334 If the -T option was specified, varnishd will offer a command-line man‐
335 agement interface on the specified address and port. The recommended
336 way of connecting to the command-line management interface is through
337 varnishadm(1).
338 The commands available are documented in varnish-cli(7).
340 CLI Command File
342 The -I option makes it possible to run arbitrary management commands
343 when varnishd is launched, before the worker process is started. In
344 particular, this is the way to load configurations, apply labels to
345 them, and make a VCL instance active that uses those labels on startup:
346 vcl.load panic /etc/varnish_panic.vcl
348 vcl.load siteA0 /etc/varnish_siteA.vcl
349 vcl.load siteB0 /etc/varnish_siteB.vcl
350 vcl.load siteC0 /etc/varnish_siteC.vcl
351 vcl.label siteA siteA0
352 vcl.label siteB siteB0
353 vcl.label siteC siteC0
354 vcl.load main /etc/varnish_main.vcl
355 vcl.use main
356 Every line in the file, including the last line, must be terminated by
358 a newline or carriage return.
359 If a command in the file is prefixed with '-', failure will not abort
361 the startup.
362
364 Run Time Parameter Flags
365 Runtime parameters are marked with shorthand flags to avoid repeating
366 the same text over and over in the table below. The meaning of the
367 flags are:
368 · experimental
370 We have no solid information about good/bad/optimal values for this
372 parameter. Feedback with experience and observations are most wel‐
373 come.
374 · delayed
376 This parameter can be changed on the fly, but will not take effect
378 immediately.
379 · restart
381 The worker process must be stopped and restarted, before this parame‐
383 ter takes effect.
384 · reload
386 The VCL programs must be reloaded for this parameter to take effect.
388 · experimental
390 We're not really sure about this parameter, tell us what you find.
392 · wizard
394 Do not touch unless you really know what you're doing.
396 · only_root
398 Only works if varnishd is running as root.
400 Default Value Exceptions on 32 bit Systems
402 Be aware that on 32 bit systems, certain default values are reduced
403 relative to the values listed below, in order to conserve VM space:
404 · workspace_client: 16k
406 · http_resp_size: 8k
408 · http_req_size: 12k
410 · gzip_stack_buffer: 4k
412 · thread_pool_stack: 64k
414 List of Parameters
416 This text is produced from the same text you will find in the CLI if
417 you use the param.show command:
418 accept_filter
420 NB: This parameter depends on a feature which is not available on all
421 platforms.
422 · Units: bool
424 · Default: off
426 · Flags:
428 Enable kernel accept-filters.
430 acceptor_sleep_decay
432 · Default: 0.9
433 · Minimum: 0
435 · Maximum: 1
437 · Flags: experimental
439 If we run out of resources, such as file descriptors or worker threads,
441 the acceptor will sleep between accepts. This parameter (multiplica‐
442 tively) reduce the sleep duration for each successful accept. (ie: 0.9
443 = reduce by 10%)
444 acceptor_sleep_incr
446 · Units: seconds
447 · Default: 0.000
449 · Minimum: 0.000
451 · Maximum: 1.000
453 · Flags: experimental
455 If we run out of resources, such as file descriptors or worker threads,
457 the acceptor will sleep between accepts. This parameter control how
458 much longer we sleep, each time we fail to accept a new connection.
459 acceptor_sleep_max
461 · Units: seconds
462 · Default: 0.050
464 · Minimum: 0.000
466 · Maximum: 10.000
468 · Flags: experimental
470 If we run out of resources, such as file descriptors or worker threads,
472 the acceptor will sleep between accepts. This parameter limits how
473 long it can sleep between attempts to accept new connections.
474 auto_restart
476 · Units: bool
477 · Default: on
479 Automatically restart the child/worker process if it dies.
481 backend_idle_timeout
483 · Units: seconds
484 · Default: 60.000
486 · Minimum: 1.000
488 Timeout before we close unused backend connections.
490 backend_local_error_holddown
492 · Units: seconds
493 · Default: 10.000
495 · Minimum: 0.000
497 · Flags: experimental
499 When connecting to backends, certain error codes (EADDRNOTAVAIL, EAC‐
501 CESS, EPERM) signal a local resource shortage or configuration issue
502 for which retrying connection attempts may worsen the situation due to
503 the complexity of the operations involved in the kernel. This parame‐
504 ter prevents repeated connection attempts for the configured duration.
505 backend_remote_error_holddown
507 · Units: seconds
508 · Default: 0.250
510 · Minimum: 0.000
512 · Flags: experimental
514 When connecting to backends, certain error codes (ECONNREFUSED, ENETUN‐
516 REACH) signal fundamental connection issues such as the backend not
517 accepting connections or routing problems for which repeated connection
518 attempts are considered useless This parameter prevents repeated con‐
519 nection attempts for the configured duration.
520 ban_cutoff
522 · Units: bans
523 · Default: 0
525 · Minimum: 0
527 · Flags: experimental
529 Expurge long tail content from the cache to keep the number of bans
531 below this value. 0 disables.
532 When this parameter is set to a non-zero value, the ban lurker contin‐
534 ues to work the ban list as usual top to bottom, but when it reaches
535 the ban_cutoff-th ban, it treats all objects as if they matched a ban
536 and expurges them from cache. As actively used objects get tested
537 against the ban list at request time and thus are likely to be associ‐
538 ated with bans near the top of the ban list, with ban_cutoff, least
539 recently accessed objects (the "long tail") are removed.
540 This parameter is a safety net to avoid bad response times due to bans
542 being tested at lookup time. Setting a cutoff trades response time for
543 cache efficiency. The recommended value is proportional to
544 rate(bans_lurker_tests_tested) / n_objects while the ban lurker is
545 working, which is the number of bans the system can sustain. The addi‐
546 tional latency due to request ban testing is in the order of ban_cutoff
547 / rate(bans_lurker_tests_tested). For example, for
548 rate(bans_lurker_tests_tested) = 2M/s and a tolerable latency of 100ms,
549 a good value for ban_cutoff may be 200K.
550 ban_dups
552 · Units: bool
553 · Default: on
555 Eliminate older identical bans when a new ban is added. This saves CPU
557 cycles by not comparing objects to identical bans. This is a waste of
558 time if you have many bans which are never identical.
559 ban_lurker_age
561 · Units: seconds
562 · Default: 60.000
564 · Minimum: 0.000
566 The ban lurker will ignore bans until they are this old. When a ban is
568 added, the active traffic will be tested against it as part of object
569 lookup. Because many applications issue bans in bursts, this parameter
570 holds the ban-lurker off until the rush is over. This should be set to
571 the approximate time which a ban-burst takes.
572 ban_lurker_batch
574 · Default: 1000
575 · Minimum: 1
577 The ban lurker sleeps ${ban_lurker_sleep} after examining this many
579 objects. Use this to pace the ban-lurker if it eats too many
580 resources.
581 ban_lurker_holdoff
583 · Units: seconds
584 · Default: 0.010
586 · Minimum: 0.000
588 · Flags: experimental
590 How long the ban lurker sleeps when giving way to lookup due to lock
592 contention.
593 ban_lurker_sleep
595 · Units: seconds
596 · Default: 0.010
598 · Minimum: 0.000
600 How long the ban lurker sleeps after examining ${ban_lurker_batch}
602 objects. Use this to pace the ban-lurker if it eats too many
603 resources. A value of zero will disable the ban lurker entirely.
604 between_bytes_timeout
606 · Units: seconds
607 · Default: 60.000
609 · Minimum: 0.000
611 We only wait for this many seconds between bytes received from the
613 backend before giving up the fetch. A value of zero means never give
614 up. VCL values, per backend or per backend request take precedence.
615 This parameter does not apply to pipe'ed requests.
616 cc_command
618 · Default: exec clang -g -O2 -Wall -Werror -Wno-error=unused-result
619 t-Werror t-Wall t-Wno-format-y2k t-W t-Wstrict-prototypes t-Wmiss‐
620 ing-prototypes t-Wpointer-arith t-Wreturn-type t-Wcast-qual
621 t-Wwrite-strings t-Wswitch t-Wshadow t-Wunused-parameter
622 t-Wcast-align t-Wchar-subscripts t-Wnested-externs t-Wextra
623 t-Wno-sign-compare -fstack-protector -Wno-missing-field-initial‐
624 izers -pthread -fpic -shared -Wl,-x -o %o %s
625 · Flags: must_reload
627 Command used for compiling the C source code to a dlopen(3) loadable
629 object. Any occurrence of %s in the string will be replaced with the
630 source file name, and %o will be replaced with the output file name.
631 cli_limit
633 · Units: bytes
634 · Default: 48k
636 · Minimum: 128b
638 · Maximum: 99999999b
640 Maximum size of CLI response. If the response exceeds this limit, the
642 response code will be 201 instead of 200 and the last line will indi‐
643 cate the truncation.
644 cli_timeout
646 · Units: seconds
647 · Default: 60.000
649 · Minimum: 0.000
651 Timeout for the childs replies to CLI requests from the mgt_param.
653 clock_skew
655 · Units: seconds
656 · Default: 10
658 · Minimum: 0
660 How much clockskew we are willing to accept between the backend and our
662 own clock.
663 clock_step
665 · Units: seconds
666 · Default: 1.000
668 · Minimum: 0.000
670 How much observed clock step we are willing to accept before we panic.
672 connect_timeout
674 · Units: seconds
675 · Default: 3.500
677 · Minimum: 0.000
679 Default connection timeout for backend connections. We only try to con‐
681 nect to the backend for this many seconds before giving up. VCL can
682 override this default value for each backend and backend request.
683 critbit_cooloff
685 · Units: seconds
686 · Default: 180.000
688 · Minimum: 60.000
690 · Maximum: 254.000
692 · Flags: wizard
694 How long the critbit hasher keeps deleted objheads on the cooloff list.
696 debug
698 · Default: none
699 Enable/Disable various kinds of debugging.
701 none Disable all debugging
703 Use +/- prefix to set/reset individual bits:
705 req_state
707 VSL Request state engine
708 workspace
710 VSL Workspace operations
711 waiter VSL Waiter internals
713 waitinglist
715 VSL Waitinglist events
716 syncvsl
718 Make VSL synchronous
719 hashedge
721 Edge cases in Hash
722 vclrel Rapid VCL release
724 lurker VSL Ban lurker
726 esi_chop
728 Chop ESI fetch to bits
729 flush_head
731 Flush after http1 head
732 vtc_mode
734 Varnishtest Mode
735 witness
737 Emit WITNESS lock records
738 vsm_keep
740 Keep the VSM file on restart
741 drop_pools
743 Drop thread pools (testing)
744 slow_acceptor
746 Slow down Acceptor
747 h2_nocheck
749 Disable various H2 checks
750 vmod_so_keep
752 Keep copied VMOD libraries
753 processors
755 Fetch/Deliver processors
756 protocol
758 Protocol debugging
759 vcl_keep
761 Keep VCL C and so files
762 default_grace
764 · Units: seconds
765 · Default: 10.000
767 · Minimum: 0.000
769 · Flags: obj_sticky
771 Default grace period. We will deliver an object this long after it has
773 expired, provided another thread is attempting to get a new copy.
774 default_keep
776 · Units: seconds
777 · Default: 0.000
779 · Minimum: 0.000
781 · Flags: obj_sticky
783 Default keep period. We will keep a useless object around this long,
785 making it available for conditional backend fetches. That means that
786 the object will be removed from the cache at the end of ttl+grace+keep.
787 default_ttl
789 · Units: seconds
790 · Default: 120.000
792 · Minimum: 0.000
794 · Flags: obj_sticky
796 The TTL assigned to objects if neither the backend nor the VCL code
798 assigns one.
799 esi_iovs
801 · Units: struct iovec
802 · Default: 10
804 · Minimum: 3
806 · Maximum: 1024
808 · Flags: wizard
810 Number of io vectors to allocate on the thread workspace for ESI
812 requests.
813 feature
815 · Default: none
816 Enable/Disable various minor features.
818 none Disable all features.
820 Use +/- prefix to enable/disable individual feature:
822 short_panic
824 Short panic message.
825 wait_silo
827 Wait for persistent silo.
828 no_coredump
830 No coredumps.
831 esi_ignore_https
833 Treat HTTPS as HTTP in ESI:includes
834 esi_disable_xml_check
836 Don't check of body looks like XML
837 esi_ignore_other_elements
839 Ignore non-esi XML-elements
840 esi_remove_bom
842 Remove UTF-8 BOM
843 https_scheme
845 Also split https URIs
846 http2 Support HTTP/2 protocol
848 http_date_postel
850 Relax parsing of timestamps in HTTP headers
851 fetch_chunksize
853 · Units: bytes
854 · Default: 16k
856 · Minimum: 4k
858 · Flags: experimental
860 The default chunksize used by fetcher. This should be bigger than the
862 majority of objects with short TTLs. Internal limits in the stor‐
863 age_file module makes increases above 128kb a dubious idea.
864 fetch_maxchunksize
866 · Units: bytes
867 · Default: 0.25G
869 · Minimum: 64k
871 · Flags: experimental
873 The maximum chunksize we attempt to allocate from storage. Making this
875 too large may cause delays and storage fragmentation.
876 first_byte_timeout
878 · Units: seconds
879 · Default: 60.000
881 · Minimum: 0.000
883 Default timeout for receiving first byte from backend. We only wait for
885 this many seconds for the first byte before giving up. A value of 0
886 means it will never time out. VCL can override this default value for
887 each backend and backend request. This parameter does not apply to
888 pipe.
889 gzip_buffer
891 · Units: bytes
892 · Default: 32k
894 · Minimum: 2k
896 · Flags: experimental
898 Size of malloc buffer used for gzip processing. These buffers are used
900 for in-transit data, for instance gunzip'ed data being sent to a
901 client.Making this space to small results in more overhead, writes to
902 sockets etc, making it too big is probably just a waste of memory.
903 gzip_level
905 · Default: 6
906 · Minimum: 0
908 · Maximum: 9
910 Gzip compression level: 0=debug, 1=fast, 9=best
912 gzip_memlevel
914 · Default: 8
915 · Minimum: 1
917 · Maximum: 9
919 Gzip memory level 1=slow/least, 9=fast/most compression. Memory impact
921 is 1=1k, 2=2k, ... 9=256k.
922 h2_header_table_size
924 · Units: bytes
925 · Default: 4k
927 · Minimum: 0b
929 HTTP2 header table size. This is the size that will be used for the
931 HPACK dynamic decoding table.
932 h2_initial_window_size
934 · Units: bytes
935 · Default: 65535b
937 · Minimum: 0b
939 · Maximum: 2147483647b
941 HTTP2 initial flow control window size.
943 h2_max_concurrent_streams
945 · Units: streams
946 · Default: 100
948 · Minimum: 0
950 HTTP2 Maximum number of concurrent streams. This is the number of
952 requests that can be active at the same time for a single HTTP2 connec‐
953 tion.
954 h2_max_frame_size
956 · Units: bytes
957 · Default: 16k
959 · Minimum: 16k
961 · Maximum: 16777215b
963 HTTP2 maximum per frame payload size we are willing to accept.
965 h2_max_header_list_size
967 · Units: bytes
968 · Default: 2147483647b
970 · Minimum: 0b
972 HTTP2 maximum size of an uncompressed header list.
974 h2_rx_window_increment
976 · Units: bytes
977 · Default: 1M
979 · Minimum: 1M
981 · Maximum: 1G
983 · Flags: wizard
985 HTTP2 Receive Window Increments. How big credits we send in WIN‐
987 DOW_UPDATE frames Only affects incoming request bodies (ie: POST, PUT
988 etc.)
989 h2_rx_window_low_water
991 · Units: bytes
992 · Default: 10M
994 · Minimum: 65535b
996 · Maximum: 1G
998 · Flags: wizard
1000 HTTP2 Receive Window low water mark. We try to keep the window at
1002 least this big Only affects incoming request bodies (ie: POST, PUT
1003 etc.)
1004 http_gzip_support
1006 · Units: bool
1007 · Default: on
1009 Enable gzip support. When enabled Varnish request compressed objects
1011 from the backend and store them compressed. If a client does not sup‐
1012 port gzip encoding Varnish will uncompress compressed objects on
1013 demand. Varnish will also rewrite the Accept-Encoding header of clients
1014 indicating support for gzip to:
1015 Accept-Encoding: gzip
1016 Clients that do not support gzip will have their Accept-Encoding header
1018 removed. For more information on how gzip is implemented please see the
1019 chapter on gzip in the Varnish reference.
1020 When gzip support is disabled the variables beresp.do_gzip and
1022 beresp.do_gunzip have no effect in VCL.
1023 http_max_hdr
1025 · Units: header lines
1026 · Default: 64
1028 · Minimum: 32
1030 · Maximum: 65535
1032 Maximum number of HTTP header lines we allow in
1034 {req|resp|bereq|beresp}.http (obj.http is autosized to the exact number
1035 of headers). Cheap, ~20 bytes, in terms of workspace memory. Note
1036 that the first line occupies five header lines.
1037 http_range_support
1039 · Units: bool
1040 · Default: on
1042 Enable support for HTTP Range headers.
1044 http_req_hdr_len
1046 · Units: bytes
1047 · Default: 8k
1049 · Minimum: 40b
1051 Maximum length of any HTTP client request header we will allow. The
1053 limit is inclusive its continuation lines.
1054 http_req_size
1056 · Units: bytes
1057 · Default: 32k
1059 · Minimum: 0.25k
1061 Maximum number of bytes of HTTP client request we will deal with. This
1063 is a limit on all bytes up to the double blank line which ends the HTTP
1064 request. The memory for the request is allocated from the client
1065 workspace (param: workspace_client) and this parameter limits how much
1066 of that the request is allowed to take up.
1067 http_resp_hdr_len
1069 · Units: bytes
1070 · Default: 8k
1072 · Minimum: 40b
1074 Maximum length of any HTTP backend response header we will allow. The
1076 limit is inclusive its continuation lines.
1077 http_resp_size
1079 · Units: bytes
1080 · Default: 32k
1082 · Minimum: 0.25k
1084 Maximum number of bytes of HTTP backend response we will deal with.
1086 This is a limit on all bytes up to the double blank line which ends the
1087 HTTP response. The memory for the response is allocated from the back‐
1088 end workspace (param: workspace_backend) and this parameter limits how
1089 much of that the response is allowed to take up.
1090 idle_send_timeout
1092 · Units: seconds
1093 · Default: 60.000
1095 · Minimum: 0.000
1097 · Flags: delayed
1099 Send timeout for individual pieces of data on client connections. May
1101 get extended if 'send_timeout' applies.
1102 When this timeout is hit, the session is closed.
1104 See the man page for setsockopt(2) under SO_SNDTIMEO for more informa‐
1106 tion.
1107 listen_depth
1109 · Units: connections
1110 · Default: 1024
1112 · Minimum: 0
1114 · Flags: must_restart
1116 Listen queue depth.
1118 lru_interval
1120 · Units: seconds
1121 · Default: 2.000
1123 · Minimum: 0.000
1125 · Flags: experimental
1127 Grace period before object moves on LRU list. Objects are only moved
1129 to the front of the LRU list if they have not been moved there already
1130 inside this timeout period. This reduces the amount of lock operations
1131 necessary for LRU list access.
1132 max_esi_depth
1134 · Units: levels
1135 · Default: 5
1137 · Minimum: 0
1139 Maximum depth of esi:include processing.
1141 max_restarts
1143 · Units: restarts
1144 · Default: 4
1146 · Minimum: 0
1148 Upper limit on how many times a request can restart.
1150 max_retries
1152 · Units: retries
1153 · Default: 4
1155 · Minimum: 0
1157 Upper limit on how many times a backend fetch can retry.
1159 nuke_limit
1161 · Units: allocations
1162 · Default: 50
1164 · Minimum: 0
1166 · Flags: experimental
1168 Maximum number of objects we attempt to nuke in order to make space for
1170 a object body.
1171 pcre_match_limit
1173 · Default: 10000
1174 · Minimum: 1
1176 The limit for the number of calls to the internal match() function in
1178 pcre_exec().
1179 (See: PCRE_EXTRA_MATCH_LIMIT in pcre docs.)
1181 This parameter limits how much CPU time regular expression matching can
1183 soak up.
1184 pcre_match_limit_recursion
1186 · Default: 20
1187 · Minimum: 1
1189 The recursion depth-limit for the internal match() function in a
1191 pcre_exec().
1192 (See: PCRE_EXTRA_MATCH_LIMIT_RECURSION in pcre docs.)
1194 This puts an upper limit on the amount of stack used by PCRE for cer‐
1196 tain classes of regular expressions.
1197 We have set the default value low in order to prevent crashes, at the
1199 cost of possible regexp matching failures.
1200 Matching failures will show up in the log as VCL_Error messages with
1202 regexp errors -27 or -21.
1203 Testcase r01576 can be useful when tuning this parameter.
1205 ping_interval
1207 · Units: seconds
1208 · Default: 3
1210 · Minimum: 0
1212 · Flags: must_restart
1214 Interval between pings from parent to child. Zero will disable pinging
1216 entirely, which makes it possible to attach a debugger to the child.
1217 pipe_timeout
1219 · Units: seconds
1220 · Default: 60.000
1222 · Minimum: 0.000
1224 Idle timeout for PIPE sessions. If nothing have been received in either
1226 direction for this many seconds, the session is closed.
1227 pool_req
1229 · Default: 10,100,10
1230 Parameters for per worker pool request memory pool. The three numbers
1232 are:
1233 min_pool
1235 minimum size of free pool.
1236 max_pool
1238 maximum size of free pool.
1239 max_age
1241 max age of free element.
1242 pool_sess
1244 · Default: 10,100,10
1245 Parameters for per worker pool session memory pool. The three numbers
1247 are:
1248 min_pool
1250 minimum size of free pool.
1251 max_pool
1253 maximum size of free pool.
1254 max_age
1256 max age of free element.
1257 pool_vbo
1259 · Default: 10,100,10
1260 Parameters for backend object fetch memory pool. The three numbers
1262 are:
1263 min_pool
1265 minimum size of free pool.
1266 max_pool
1268 maximum size of free pool.
1269 max_age
1271 max age of free element.
1272 prefer_ipv6
1274 · Units: bool
1275 · Default: off
1277 Prefer IPv6 address when connecting to backends which have both IPv4
1279 and IPv6 addresses.
1280 rush_exponent
1282 · Units: requests per request
1283 · Default: 3
1285 · Minimum: 2
1287 · Flags: experimental
1289 How many parked request we start for each completed request on the
1291 object. NB: Even with the implict delay of delivery, this parameter
1292 controls an exponential increase in number of worker threads.
1293 send_timeout
1295 · Units: seconds
1296 · Default: 600.000
1298 · Minimum: 0.000
1300 · Flags: delayed
1302 Total timeout for ordinary HTTP1 responses. Does not apply to some
1304 internally generated errors and pipe mode.
1305 When 'idle_send_timeout' is hit while sending an HTTP1 response, the
1307 timeout is extended unless the total time already taken for sending the
1308 response in its entirety exceeds this many seconds.
1309 When this timeout is hit, the session is closed
1311 shm_reclen
1313 · Units: bytes
1314 · Default: 255b
1316 · Minimum: 16b
1318 · Maximum: 4084
1320 Old name for vsl_reclen, use that instead.
1322 shortlived
1324 · Units: seconds
1325 · Default: 10.000
1327 · Minimum: 0.000
1329 Objects created with (ttl+grace+keep) shorter than this are always put
1331 in transient storage.
1332 sigsegv_handler
1334 · Units: bool
1335 · Default: on
1337 · Flags: must_restart
1339 Install a signal handler which tries to dump debug information on seg‐
1341 mentation faults, bus errors and abort signals.
1342 syslog_cli_traffic
1344 · Units: bool
1345 · Default: on
1347 Log all CLI traffic to syslog(LOG_INFO).
1349 tcp_fastopen
1351 · Units: bool
1352 · Default: off
1354 · Flags: must_restart
1356 Enable TCP Fast Open extension.
1358 tcp_keepalive_intvl
1360 · Units: seconds
1361 · Default: 75.000
1363 · Minimum: 1.000
1365 · Maximum: 100.000
1367 · Flags: experimental
1369 The number of seconds between TCP keep-alive probes. Ignored for Unix
1371 domain sockets.
1372 tcp_keepalive_probes
1374 · Units: probes
1375 · Default: 9
1377 · Minimum: 1
1379 · Maximum: 100
1381 · Flags: experimental
1383 The maximum number of TCP keep-alive probes to send before giving up
1385 and killing the connection if no response is obtained from the other
1386 end. Ignored for Unix domain sockets.
1387 tcp_keepalive_time
1389 · Units: seconds
1390 · Default: 7200.000
1392 · Minimum: 1.000
1394 · Maximum: 7200.000
1396 · Flags: experimental
1398 The number of seconds a connection needs to be idle before TCP begins
1400 sending out keep-alive probes. Ignored for Unix domain sockets.
1401 thread_pool_add_delay
1403 · Units: seconds
1404 · Default: 0.000
1406 · Minimum: 0.000
1408 · Flags: experimental
1410 Wait at least this long after creating a thread.
1412 Some (buggy) systems may need a short (sub-second) delay between creat‐
1414 ing threads. Set this to a few milliseconds if you see the
1415 'threads_failed' counter grow too much.
1416 Setting this too high results in insufficient worker threads.
1418 thread_pool_destroy_delay
1420 · Units: seconds
1421 · Default: 1.000
1423 · Minimum: 0.010
1425 · Flags: delayed, experimental
1427 Wait this long after destroying a thread.
1429 This controls the decay of thread pools when idle(-ish).
1431 thread_pool_fail_delay
1433 · Units: seconds
1434 · Default: 0.200
1436 · Minimum: 0.010
1438 · Flags: experimental
1440 Wait at least this long after a failed thread creation before trying to
1442 create another thread.
1443 Failure to create a worker thread is often a sign that the end is
1445 near, because the process is running out of some resource. This delay
1446 tries to not rush the end on needlessly.
1447 If thread creation failures are a problem, check that thread_pool_max
1449 is not too high.
1450 It may also help to increase thread_pool_timeout and thread_pool_min,
1452 to reduce the rate at which treads are destroyed and later recreated.
1453 thread_pool_max
1455 · Units: threads
1456 · Default: 5000
1458 · Minimum: 100
1460 · Flags: delayed
1462 The maximum number of worker threads in each pool. The minimum value
1464 depends on thread_pool_min.
1465 Do not set this higher than you have to, since excess worker threads
1467 soak up RAM and CPU and generally just get in the way of getting work
1468 done.
1469 thread_pool_min
1471 · Units: threads
1472 · Default: 100
1474 · Maximum: 5000
1476 · Flags: delayed
1478 The minimum number of worker threads in each pool. The maximum value
1480 depends on thread_pool_max.
1481 Increasing this may help ramp up faster from low load situations or
1483 when threads have expired.
1484 Minimum is 10 threads.
1486 thread_pool_reserve
1488 · Units: threads
1489 · Default: 0
1491 · Maximum: 95
1493 · Flags: delayed
1495 The number of worker threads reserved for vital tasks in each pool.
1497 Tasks may require other tasks to complete (for example, client requests
1499 may require backend requests). This reserve is to ensure that such
1500 tasks still get to run even under high load.
1501 Increasing the reserve may help setups with a high number of backend
1503 requests at the expense of client performance. Setting it too high will
1504 waste resources by keeping threads unused.
1505 Default is 0 to auto-tune (currently 5% of thread_pool_min). Minimum
1507 is 1 otherwise, maximum is 95% of thread_pool_min.
1508 thread_pool_stack
1510 · Units: bytes
1511 · Default: 48k
1513 · Minimum: 16k
1515 · Flags: delayed
1517 Worker thread stack size. This will likely be rounded up to a multiple
1519 of 4k (or whatever the page_size might be) by the kernel.
1520 The required stack size is primarily driven by the depth of the
1522 call-tree. The most common relevant determining factors in varnish core
1523 code are GZIP (un)compression, ESI processing and regular expression
1524 matches. VMODs may also require significant amounts of additional
1525 stack. The nesting depth of VCL subs is another factor, although typi‐
1526 cally not predominant.
1527 The stack size is per thread, so the maximum total memory required for
1529 worker thread stacks is in the order of size = thread_pools x
1530 thread_pool_max x thread_pool_stack.
1531 Thus, in particular for setups with many threads, keeping the stack
1533 size at a minimum helps reduce the amount of memory required by Var‐
1534 nish.
1535 On the other hand, thread_pool_stack must be large enough under all
1537 circumstances, otherwise varnish will crash due to a stack overflow.
1538 Usually, a stack overflow manifests itself as a segmentation fault (aka
1539 segfault / SIGSEGV) with the faulting address being near the stack
1540 pointer (sp).
1541 Unless stack usage can be reduced, thread_pool_stack must be increased
1543 when a stack overflow occurs. Setting it in 150%-200% increments is
1544 recommended until stack overflows cease to occur.
1545 thread_pool_timeout
1547 · Units: seconds
1548 · Default: 300.000
1550 · Minimum: 10.000
1552 · Flags: delayed, experimental
1554 Thread idle threshold.
1556 Threads in excess of thread_pool_min, which have been idle for at least
1558 this long, will be destroyed.
1559 thread_pool_watchdog
1561 · Units: seconds
1562 · Default: 60.000
1564 · Minimum: 0.100
1566 · Flags: experimental
1568 Thread queue stuck watchdog.
1570 If no queued work have been released for this long, the worker process
1572 panics itself.
1573 thread_pools
1575 · Units: pools
1576 · Default: 2
1578 · Minimum: 1
1580 · Maximum: 32
1582 · Flags: delayed, experimental
1584 Number of worker thread pools.
1586 Increasing the number of worker pools decreases lock contention. Each
1588 worker pool also has a thread accepting new connections, so for very
1589 high rates of incoming new connections on systems with many cores,
1590 increasing the worker pools may be required.
1591 Too many pools waste CPU and RAM resources, and more than one pool for
1593 each CPU is most likely detrimental to performance.
1594 Can be increased on the fly, but decreases require a restart to take
1596 effect.
1597 thread_queue_limit
1599 · Default: 20
1600 · Minimum: 0
1602 · Flags: experimental
1604 Permitted request queue length per thread-pool.
1606 This sets the number of requests we will queue, waiting for an avail‐
1608 able thread. Above this limit sessions will be dropped instead of
1609 queued.
1610 thread_stats_rate
1612 · Units: requests
1613 · Default: 10
1615 · Minimum: 0
1617 · Flags: experimental
1619 Worker threads accumulate statistics, and dump these into the global
1621 stats counters if the lock is free when they finish a job
1622 (request/fetch etc.) This parameters defines the maximum number of
1623 jobs a worker thread may handle, before it is forced to dump its accu‐
1624 mulated stats into the global counters.
1625 timeout_idle
1627 · Units: seconds
1628 · Default: 5.000
1630 · Minimum: 0.000
1632 Idle timeout for client connections.
1634 A connection is considered idle until we have received the full request
1636 headers.
1637 This parameter is particularly relevant for HTTP1 keepalive connec‐
1639 tions which are closed unless the next request is received before this
1640 timeout is reached.
1641 timeout_linger
1643 · Units: seconds
1644 · Default: 0.050
1646 · Minimum: 0.000
1648 · Flags: experimental
1650 How long the worker thread lingers on an idle session before handing it
1652 over to the waiter. When sessions are reused, as much as half of all
1653 reuses happen within the first 100 msec of the previous request com‐
1654 pleting. Setting this too high results in worker threads not doing
1655 anything for their keep, setting it too low just means that more ses‐
1656 sions take a detour around the waiter.
1657 vcc_allow_inline_c
1659 · Units: bool
1660 · Default: off
1662 Allow inline C code in VCL.
1664 vcc_err_unref
1666 · Units: bool
1667 · Default: on
1669 Unreferenced VCL objects result in error.
1671 vcc_unsafe_path
1673 · Units: bool
1674 · Default: on
1676 Allow '/' in vmod & include paths. Allow 'import ... from ...'.
1678 vcl_cooldown
1680 · Units: seconds
1681 · Default: 600.000
1683 · Minimum: 0.000
1685 How long a VCL is kept warm after being replaced as the active VCL
1687 (granularity approximately 30 seconds).
1688 vcl_dir
1690 · Default: /opt/varnish/etc/varnish:/opt/varnish/share/varnish/vcl
1691 Old name for vcl_path, use that instead.
1693 vcl_path
1695 · Default: /opt/varnish/etc/varnish:/opt/varnish/share/varnish/vcl
1696 Directory (or colon separated list of directories) from which relative
1698 VCL filenames (vcl.load and include) are to be found. By default Var‐
1699 nish searches VCL files in both the system configuration and shared
1700 data directories to allow packages to drop their VCL files in a stan‐
1701 dard location where relative includes would work.
1702 vmod_dir
1704 · Default: /opt/varnish/lib/varnish/vmods
1705 Old name for vmod_path, use that instead.
1707 vmod_path
1709 · Default: /opt/varnish/lib/varnish/vmods
1710 Directory (or colon separated list of directories) where VMODs are to
1712 be found.
1713 vsl_buffer
1715 · Units: bytes
1716 · Default: 4k
1718 · Minimum: 267
1720 Bytes of (req-/backend-)workspace dedicated to buffering VSL records.
1722 When this parameter is adjusted, most likely workspace_client and
1723 workspace_backend will have to be adjusted by the same amount.
1724 Setting this too high costs memory, setting it too low will cause more
1726 VSL flushes and likely increase lock-contention on the VSL mutex.
1727 The minimum tracks the vsl_reclen parameter + 12 bytes.
1729 vsl_mask
1731 · Default: -ObjProtocol,-ObjStatus,-ObjReason,-Obj‐
1732 Header,-VCL_trace,-Work‐
1733 Thread,-Hash,-VfpAcct,-H2RxHdr,-H2RxBody,-H2TxHdr,-H2TxBody
1734 Mask individual VSL messages from being logged.
1736 default
1738 Set default value
1739 Use +/- prefix in front of VSL tag name, to mask/unmask individual VSL
1741 messages.
1742 vsl_reclen
1744 · Units: bytes
1745 · Default: 255b
1747 · Minimum: 16b
1749 · Maximum: 4084b
1751 Maximum number of bytes in SHM log record.
1753 The maximum tracks the vsl_buffer parameter - 12 bytes.
1755 vsl_space
1757 · Units: bytes
1758 · Default: 80M
1760 · Minimum: 1M
1762 · Maximum: 4G
1764 · Flags: must_restart
1766 The amount of space to allocate for the VSL fifo buffer in the VSM mem‐
1768 ory segment. If you make this too small, varnish{ncsa|log} etc will
1769 not be able to keep up. Making it too large just costs memory
1770 resources.
1771 vsm_free_cooldown
1773 · Units: seconds
1774 · Default: 60.000
1776 · Minimum: 10.000
1778 · Maximum: 600.000
1780 How long VSM memory is kept warm after a deallocation (granularity
1782 approximately 2 seconds).
1783 vsm_space
1785 · Units: bytes
1786 · Default: 1M
1788 · Minimum: 1M
1790 · Maximum: 1G
1792 DEPRECATED: This parameter is ignored. There is no global limit on
1794 amount of shared memory now.
1795 workspace_backend
1797 · Units: bytes
1798 · Default: 64k
1800 · Minimum: 1k
1802 · Flags: delayed
1804 Bytes of HTTP protocol workspace for backend HTTP req/resp. If larger
1806 than 4k, use a multiple of 4k for VM efficiency.
1807 workspace_client
1809 · Units: bytes
1810 · Default: 64k
1812 · Minimum: 9k
1814 · Flags: delayed
1816 Bytes of HTTP protocol workspace for clients HTTP req/resp. Use a mul‐
1818 tiple of 4k for VM efficiency. For HTTP/2 compliance this must be at
1819 least 20k, in order to receive fullsize (=16k) frames from the client.
1820 That usually happens only in POST/PUT bodies. For other traffic-pat‐
1821 terns smaller values work just fine.
1822 workspace_session
1824 · Units: bytes
1825 · Default: 0.50k
1827 · Minimum: 0.25k
1829 · Flags: delayed
1831 Allocation size for session structure and workspace. The workspace
1833 is primarily used for TCP connection addresses. If larger than 4k, use
1834 a multiple of 4k for VM efficiency.
1835 workspace_thread
1837 · Units: bytes
1838 · Default: 2k
1840 · Minimum: 0.25k
1842 · Maximum: 8k
1844 · Flags: delayed
1846 Bytes of auxiliary workspace per thread. This workspace is used for
1848 certain temporary data structures during the operation of a worker
1849 thread. One use is for the IO-vectors used during delivery. Setting
1850 this parameter too low may increase the number of writev() syscalls,
1851 setting it too high just wastes space. ~0.1k + UIO_MAXIOV *
1852 sizeof(struct iovec) (typically = ~16k for 64bit) is considered the
1853 maximum sensible value under any known circumstances (excluding exotic
1854 vmod use).
1855
1857 Varnish and bundled tools will, in most cases, exit with one of the
1858 following codes
1859 · 0 OK
1861 · 1 Some error which could be system-dependent and/or transient
1863 · 2 Serious configuration / parameter error - retrying with the same
1865 configuration / parameters is most likely useless
1866 The varnishd master process may also OR its exit code
1868 · with 0x20 when the varnishd child process died,
1870 · with 0x40 when the varnishd child process was terminated by a signal
1872 and
1873 · with 0x80 when a core was dumped.
1875
1877 · varnishlog(1)
1878 · varnishhist(1)
1880 · varnishncsa(1)
1882 · varnishstat(1)
1884 · varnishtop(1)
1886 · varnish-cli(7)
1888 · vcl(7)
1890
1892 The varnishd daemon was developed by Poul-Henning Kamp in cooperation
1893 with Verdens Gang AS and Varnish Software.
1894 This manual page was written by Dag-Erling Smørgrav with updates by
1896 Stig Sandbeck Mathisen <ssm@debian.org>, Nils Goroll and others.
1897
1899 This document is licensed under the same licence as Varnish itself. See
1900 LICENCE for details.
1901 · Copyright (c) 2007-2015 Varnish Software AS
1903 VARNISHD(1)