mirrors: include timestamp in logs
[girocco/readme.git] / jobd / update.sh
blobc51d30286b62212994a32927f8f5fef451e6271f
1 #!/bin/sh
3 . @basedir@/shlib.sh
5 set -e
7 if [ $# -ne 1 ]; then
8 echo "Usage: update.sh projname" >&2
9 exit 1
12 # date -R is linux-only, POSIX equivalent is '+%a, %d %b %Y %T %z'
13 datefmt='+%a, %d %b %Y %T %z'
15 git_fetch_q_progress() {
16 PATH="$var_git_exec_path:$cfg_basedir/bin:$PATH" @basedir@/jobd/git-fetch-q-progress.sh "$@"
19 # freshen_loose_objects full-sha ...
20 # if "$n" is a loose object, set its modification time to now
21 # otherwise silently do nothing with no error. To facilitate conversion
22 # of mirror projects to push projects we also add group write permission.
23 freshen_loose_objects() {
24 _list=
25 for _sha; do
26 _fn="${_sha#??}"
27 _shard="${_sha%$_fn}"
28 _list="$_list objects/$_shard/$_fn"
29 done
30 if [ -n "$_list" ]; then
31 chmod ug+w $_list 2>/dev/null || :
32 touch -c $_list 2>/dev/null || :
36 # darcs fast-export | git fast-import with error handling
37 git_darcs_fetch() (
38 set_utf8_locale
39 _err1=
40 _err2=
41 exec 3>&1
42 { read -r _err1 || :; read -r _err2 || :; } <<-EOT
44 exec 4>&3 3>&1 1>&4 4>&-
46 _e1=0
47 "$cfg_basedir"/bin/darcs-fast-export \
48 --export-marks="$(pwd)/dfe-marks" \
49 --import-marks="$(pwd)/dfe-marks" "$1" 3>&- || _e1=$?
50 echo $_e1 >&3
51 } |
53 _e2=0
54 git fast-import \
55 --export-marks="$(pwd)/gfi-marks" \
56 --export-pack-edges="$(pwd)/gfi-packs" \
57 --import-marks="$(pwd)/gfi-marks" \
58 --force 3>&- || _e2=$?
59 echo $_e2 >&3
62 EOT
63 exec 3>&-
64 [ "$_err1" = 0 ] && [ "$_err2" = 0 ]
65 return $?
68 # bzr fast-export | git fast-import with error handling
69 git_bzr_fetch() (
70 set_utf8_locale
71 BZR_LOG=/dev/null
72 export BZR_LOG
73 _err1=
74 _err2=
75 exec 3>&1
76 { read -r _err1 || :; read -r _err2 || :; } <<-EOT
78 exec 4>&3 3>&1 1>&4 4>&-
80 _e1=0
81 bzr fast-export --plain \
82 --export-marks="$(pwd)/bfe-marks" \
83 --import-marks="$(pwd)/bfe-marks" "$1" 3>&- || _e1=$?
84 echo $_e1 >&3
85 } |
87 _e2=0
88 git fast-import \
89 --export-marks="$(pwd)/gfi-marks" \
90 --export-pack-edges="$(pwd)/gfi-packs" \
91 --import-marks="$(pwd)/gfi-marks" \
92 --force 3>&- || _e2=$?
93 echo $_e2 >&3
96 EOT
97 exec 3>&-
98 [ "$_err1" = 0 ] && [ "$_err2" = 0 ]
99 return $?
102 [ -n "$cfg_mirror" ] || { echo "Mirroring is disabled" >&2; exit 0; }
104 umask 002
105 [ "$cfg_permission_control" != "Hooks" ] || umask 000
106 clean_git_env
108 proj="${1%.git}"
109 cd "$cfg_reporoot/$proj.git"
111 # Activate a mini-gc if needed
112 # We do this here as well as after a successful fetch so that if we're stuck
113 # in a fetch loop where fetches are succeeding in fetching new packs but the
114 # ref update is failing for some reason (perhaps a non-commit under refs/heads)
115 # and a previous invokation therefore had a "bang" exit then we will still
116 # get the .needsgc flag set in a timely fashion to avoid excess pack build up.
117 check_and_set_needsgc
119 trap 'if [ $? != 0 ]; then echo "update failed dir: $PWD" >&2; fi; rm -f "$bang_log"' EXIT
120 trap 'exit 130' INT
121 trap 'exit 143' TERM
123 if [ "${force_update:-0}" = "0" ] && check_interval lastrefresh $cfg_min_mirror_interval; then
124 progress "= [$proj] update skip (last at $(config_get lastrefresh))"
125 exit 0
127 if [ -e .nofetch ]; then
128 progress "x [$proj] update disabled (.nofetch exists)"
129 exit 0
131 progress "+ [$proj] update ($(date))"
133 # Any pre-existing FETCH_HEAD from a previous fetch, failed or not, is garbage
134 rm -f FETCH_HEAD
136 # A previous failed update attempt can leave a huge tmp_pack_XXXXXX file behind.
137 # Since no pushes are allowed to mirrors, we know that any such files that exist
138 # at this point in time are garbage and can be safely deleted, we do not even
139 # need to check how old they are. A tmp_idx_XXXXXX file is also created during
140 # the later stages of the fetch process, so we kill any of those as well.
141 find -L objects/pack -maxdepth 1 -type f -name "tmp_pack_?*" -exec rm -f '{}' + || :
142 find -L objects/pack -maxdepth 1 -type f -name "tmp_idx_?*" -exec rm -f '{}' + || :
144 # Make sure we have a reflogs subdirectory and abort the update if not
145 # This should not count as a normal "bang" failure if unsuccessful
146 [ -d reflogs ] || mkdir -p reflogs >/dev/null 2>&1 || :
147 [ -d reflogs ]
149 bang_setup
150 bang_action="update"
151 bang_trap() {
152 if [ -n "$1" ]; then
153 # Throttle retries
154 # Since gitweb shows the .last_refresh date, it's safe to update
155 # gitweb.lastrefresh to throttle the updates w/o corrupting the
156 # last refresh date display on the gitweb summary page
157 # It's therefore important that we do NOT touch .last_refresh here
158 config_set lastrefresh "$(date "$datefmt")"
162 bang echo "Project: $proj"
163 bang echo " Date: $(TZ=UTC date '+%Y-%m-%d %T UTC')"
164 bang echo ""
165 mail="$(config_get owner)" || :
166 url="$(config_get baseurl)" || :
167 case "$url" in *" "*|*" "*|"")
168 bang_eval 'echo "Bad mirror URL (\"$url\")"; ! :'
169 exit 1
170 esac
171 bang echo "Mirroring from URL \"$url\""
172 bang echo ""
173 statusok="$(git config --bool gitweb.statusupdates 2>/dev/null || echo true)"
174 mailaddrs=
175 [ "$statusok" = "false" ] || [ -z "$mail" ] || mailaddrs="$mail"
176 [ -z "$cfg_admincc" ] || [ "$cfg_admincc" = "0" ] || [ -z "$cfg_admin" ] ||
177 if [ -z "$mailaddrs" ]; then mailaddrs="$cfg_admin"; else mailaddrs="$mailaddrs,$cfg_admin"; fi
179 bang_eval "git for-each-ref --format '%(refname) %(objectname)' >.refs-temp"
180 bang_eval "LC_ALL=C sort -b -k1,1 <.refs-temp >.refs-before"
181 ! [ -e .delaygc ] || >.allowgc || :
182 svnpackcreated=
184 # Make sure we don't get any unwanted loose objects
185 # Starting with Git v2.10.0 fast-import can generate loose objects unless we
186 # tweak its configuration to prevent that
187 git_add_config 'fetch.unpackLimit=1'
188 # Note the git config documentation is wrong
189 # transfer.unpackLimit, if set, overrides fetch.unpackLimit
190 git_add_config 'transfer.unpackLimit=1'
191 # But not the Git v2.10.0 and later fastimport.unpackLimit which improperly uses <= instead of <
192 git_add_config 'fastimport.unpackLimit=0'
194 case "$url" in
195 svn://* | svn+http://* | svn+https://* | svn+file://* | svn+ssh://*)
196 [ -n "$cfg_mirror_svn" ] || { echo "Mirroring svn is disabled" >&2; exit 0; }
197 # Use an 'anonsvn' username as is commonly used for anonymous svn
198 # Use an 'anonsvn' password as is commonly used for anonymous svn
199 GIT_ASKPASS_PASSWORD=anonsvn
200 export GIT_ASKPASS_PASSWORD
201 # Update the git svn url to match baseurl but be cognizant of any
202 # needed prefix changes. See the comments in taskd/clone.sh about
203 # why we need to put up with a prefix in the first place.
204 case "$url" in svn+ssh://*) svnurl="$url";; *) svnurl="${url#svn+}";; esac
205 svnurl="${svnurl%/}"
206 svnurlold="$(config_get svnurl)" || :
207 if [ "$svnurl" != "$svnurlold" ]; then
208 # We better already have an svn-remote.svn.fetch setting
209 bang test -n "$(git config --get-all svn-remote.svn.fetch)" || :
210 # the only way to truly know what the proper prefix is
211 # is to attempt a fresh git-svn init -s on the new url
212 rm -rf svn-new-url || :
213 # We require svn info to succeed on the URL otherwise it's
214 # simply not a valid URL and without using -s on the init it
215 # will not otherwise be tested until the fetch
216 bang eval 'svn --non-interactive --username anonsvn --password anonsvn info "$svnurl" >/dev/null'
217 bang mkdir svn-new-url
218 GIT_DIR=svn-new-url bang git init --bare --quiet
219 # We initially use -s for the init which will possibly shorten
220 # the URL. However, the shortening can fail if a password is
221 # not required for the longer version but is for the shorter,
222 # so try again without -s if the -s version fails.
223 cmdstr='git svn init --username=anonsvn --prefix "" -s "$svnurl" </dev/null >/dev/null 2>&1 || '
224 cmdstr="$cmdstr"'git svn init --username=anonsvn --prefix "" "$svnurl" </dev/null >/dev/null 2>&1'
225 GIT_DIR=svn-new-url bang eval "$cmdstr"
226 gitsvnurl="$(GIT_DIR=svn-new-url git config --get svn-remote.svn.url)" || :
227 gitsvnfetch="$(GIT_DIR=svn-new-url git config --get svn-remote.svn.fetch)" || :
228 gitsvnprefixnew="${gitsvnfetch%%:*}"
229 gitsvnsuffixnew="${gitsvnprefixnew##*/}"
230 gitsvnprefixnew="${gitsvnprefixnew%$gitsvnsuffixnew}"
231 rm -rf svn-new-url || :
232 # Using GIT_DIR= with bang leaves it set to svn-new-url, so reset it to .
233 GIT_DIR=.
234 if [ "$gitsvnurl" != "$(git config --get svn-remote.svn.url || :)" ]; then
235 # The url has been changed.
236 # We must update the url and replace the prefix on all config items
237 gitsvnfetch="$(git config --get-all svn-remote.svn.fetch | head -1)" || :
238 gitsvnprefixold="${gitsvnfetch%%:*}"
239 gitsvnsuffixold="${gitsvnprefixold##*/}"
240 gitsvnprefixold="${gitsvnprefixold%$gitsvnsuffixold}"
241 git config --remove-section 'svn-remote.svnnew' 2>/dev/null || :
242 git config 'svn-remote.svnnew.url' "$gitsvnurl"
243 git config --get-regexp '^svn-remote\.svn\.' |
244 while read -r sname sval; do
245 case "$sname" in
246 svn-remote.svn.fetch|svn-remote.svn.branches|svn-remote.svn.tags)
247 sname="${sname#svn-remote.svn.}"
248 sval="${sval#$gitsvnprefixold}"
249 bang git config --add "svn-remote.svnnew.$sname" "${gitsvnprefixnew}$sval"
250 esac
251 done
252 test $? -eq 0
253 bang git config -f svn/.metadata svn-remote.svn.reposRoot "$gitsvnurl"
254 bang git config --remove-section svn-remote.svn
255 bang git config --rename-section svn-remote.svnnew svn-remote.svn
257 bang config_set svnurl "$svnurl"
259 # remove any stale *.lock files greater than 1 hour old in case
260 # git-svn was killed on the last update because it took too long
261 find -L svn -type f -name '*.lock' -mmin +60 -exec rm -f '{}' + 2>/dev/null || :
262 # remember the starting time so we can easily combine fetched loose objects
263 # we sleep for 1 second after creating .svnpack to make sure all objects are newer
264 if ! [ -e .svnpack ]; then
265 svnpackcreated=1
266 rm -f .svnpack
267 >.svnpack
268 sleep 1
270 GIT_DIR=. bang git svn fetch --log-window-size=$var_log_window_size --username=anonsvn --quiet </dev/null
271 # git svn does not preserve group permissions in the svn subdirectory
272 chmod -R ug+rw,o+r svn
273 # git svn also leaves behind ref turds that end with @nnn
274 # We get rid of them now
275 git for-each-ref --format='%(refname)' |
276 LC_ALL=C sed '/^..*@[1-9][0-9]*$/!d; s/^/delete /' |
277 git_updateref_stdin
278 unset GIT_ASKPASS_PASSWORD
280 darcs://*)
281 [ -n "$cfg_mirror_darcs" ] || { echo "Mirroring darcs is disabled" >&2; exit 0; }
282 httpurl="http://${url#darcs://}"
283 # remove any stale lock files greater than 1 hour old in case
284 # darcs_fast_export was killed on the last update because it took too long
285 find -L *.darcs -maxdepth 2 -type f -name 'lock' -mmin +60 -exec rm -f '{}' + 2>/dev/null || :
286 bang git_darcs_fetch "$httpurl"
288 bzr://*)
289 [ -n "$cfg_mirror_bzr" ] || { echo "Mirroring bzr is disabled" >&2; exit 0; }
290 bzrurl="${url#bzr://}"
291 bang git_bzr_fetch "$bzrurl"
293 hg+http://* | hg+https://* | hg+file://* | hg+ssh://*)
294 [ -n "$cfg_mirror_hg" ] || { echo "Mirroring hg is disabled" >&2; exit 0; }
295 # We just remove hg+ here, so hg+http://... becomes http://...
296 hgurl="${url#hg+}"
297 # Fetch any new updates
298 bang hg -R "$(pwd)/repo.hg" pull
299 # Do the fast-export | fast-import
300 bang git_hg_fetch
303 [ "$url" = "$(git config --get remote.origin.url || :)" ] || bang config_set_raw remote.origin.url "$url"
304 pruneopt=--prune
305 [ "$(git config --bool fetch.prune 2>/dev/null || :)" != "false" ] || pruneopt=
306 if ! is_gfi_mirror_url "$url"; then
307 lastwasclean=
308 [ "$(git config --bool girocco.lastupdateclean 2>/dev/null || :)" != "true" ] || lastwasclean=1
309 nextisclean=
310 [ "$(git config --bool girocco.cleanmirror 2>/dev/null || :)" != "true" ] || nextisclean=1
311 if [ "$nextisclean" != "$lastwasclean" ]; then
312 if [ -n "$nextisclean" ]; then
313 git config --replace-all remote.origin.fetch "+refs/heads/*:refs/heads/*"
314 git config --add remote.origin.fetch "+refs/tags/*:refs/tags/*"
315 git config --add remote.origin.fetch "+refs/notes/*:refs/notes/*"
316 git config --add remote.origin.fetch "+refs/top-bases/*:refs/top-bases/*"
317 else
318 git config --replace-all remote.origin.fetch "+refs/*:refs/*"
322 # remember the starting time so we can easily detect new packs for fast-import mirrors
323 # we sleep for 1 second after creating .gfipack to make sure all packs are newer
324 if is_gfi_mirror_url "$url" && [ ! -e .gfipack ]; then
325 rm -f .gfipack
326 >.gfipack
327 sleep 1
329 fetcharg="default"
330 git config remotes.default >/dev/null 2>&1 || fetcharg="--all"
331 fetchcmd="git fetch"
332 if [ -n "$var_have_git_171" ] && [ "${show_progress:-0}" != "0" ]; then
333 # git fetch learned --progress in v1.7.1
334 case "$show_progress" in
335 [2-9]*|1[0-9]*)
336 # full volume progress with all the spammy noise
337 fetchcmd="$fetchcmd --progress"
340 # a kinder, gentler progress that doesn't leave one
341 # covered all over in exploded bits of spam afterwards
342 fetchcmd="git_fetch_q_progress"
344 esac
346 GIT_SSL_NO_VERIFY=1 bang_eval "$fetchcmd" $pruneopt --multiple "$fetcharg"
347 if ! is_gfi_mirror_url "$url" && [ "$nextisclean" != "$lastwasclean" ]; then
348 if [ -n "$nextisclean" ]; then
349 # We must manually purge the unclean refs now as even prune won't do it
350 git for-each-ref --format='%(refname)' |
351 LC_ALL=C sed \
352 -e '/^refs\/heads\//d' \
353 -e '/^refs\/tags\//d' \
354 -e '/^refs\/notes\//d' \
355 -e '/^refs\/top-bases\//d' \
356 -e 's/^/delete /' |
357 git_updateref_stdin
359 git config --bool girocco.lastupdateclean ${nextisclean:-0}
361 if [ -e .gfipack ] && is_gfi_mirror_url "$url"; then
362 find -L objects/pack -type f -newer .gfipack -name "pack-$octet20*.pack" -print >>gfi-packs
363 rm -f .gfipack
366 esac
368 # The objects subdirectories permissions must be updated now.
369 # In the case of a dumb http clone, the permissions will not be correct
370 # (missing group write) despite the core.sharedrepository=1 setting!
371 # The objects themselves seem to have the correct permissions.
372 # This problem appears to have been fixed in the most recent git versions.
373 perms=g+w
374 [ "$cfg_permission_control" != "Hooks" ] || perms=go+w
375 chmod $perms $(find -L objects -maxdepth 1 -type d) 2>/dev/null || :
377 bang git update-server-info
379 # We maintain the last refresh date in two places deliberately
380 # so that it's available as part of the config data and also
381 # as a standalone file timestamp that can be accessed without git.
382 bang config_set lastrefresh "$(date "$datefmt")"
383 { >.last_refresh; } 2>/dev/null || :
385 # Check to see if any refs changed
386 bang_eval "git for-each-ref --format '%(refname) %(objectname)' >.refs-temp"
387 bang_eval "LC_ALL=C sort -b -k1,1 <.refs-temp >.refs-after"
388 refschanged=
389 cmp -s .refs-before .refs-after || refschanged=1
391 # There's no way to get git svn to fetch packs, so we always need to run
392 # a mini-gc afterwards if svn actually fetched anything, but that's okay
393 # because it will be very quick
394 if [ -z "$refschanged" ] && [ -n "$svnpackcreated" ]; then
395 # we created the .svnpack but didn't actually fetch anything
396 # so remove it to avoid forcing a mini-gc if not necessary
397 rm -f .svnpack
399 if [ -e .svnpack ] && ! [ -e .needsgc ]; then
400 >.needsgc
403 # Force a mini-gc if $Girocco::Config::delay_gfi_redelta is false and there's
404 # at least one gfi pack present now
405 if [ -z "$cfg_delay_gfi_redelta" ] && ! [ -e .needsgc ] &&
406 [ -f gfi-packs ] && [ -s gfi-packs ] && is_gfi_mirror_url "$url"; then
407 >.needsgc
410 # Activate a mini-gc if needed
411 check_and_set_needsgc
413 # Look at which refs changed and trigger ref-change for these
414 sockpath="$cfg_chroot/etc/taskd.socket"
415 if [ -n "$refschanged" ]; then
416 bang config_set lastreceive "$(date '+%a, %d %b %Y %T %z')"
417 # We always use UTC for the log timestamp so that chroot and non-chroot match up.
418 # We don't have to worry about multiple log files since only one update runs
419 lognamets="$(TZ=UTC date '+%Y%m%d_%H%M%S')"
420 loghhmmss="${lognamets##*_}"
421 logname="reflogs/${lognamets%%_*}"
422 # We freshen the mod time to now on any old or new ref that is a loose object
423 # For old refs we do it so we will be able to keep them around for 1 day
424 # For new refs we do it in case we are about to run gc and the new ref
425 # actually points to an oldish loose object that had been unreachable
426 # We probably do not need to do it for new refs as Git tries to do that,
427 # but since we're already doing it for old refs (which Git does not do),
428 # it's almost no extra work for new refs, just in case.
430 echo "ref-changes %$proj% $proj"
431 LC_ALL=C join .refs-before .refs-after |
432 LC_ALL=C sed -e '/^[^ ][^ ]* \([^ ][^ ]*\) \1$/d' |
433 while read ref old new; do
434 echo "$loghhmmss $old $new $ref" >&3
435 freshen_loose_objects "$old" "$new"
436 echo "$old $new $ref"
437 done
438 LC_ALL=C join -v 1 .refs-before .refs-after |
439 while read ref old; do
440 echo "$loghhmmss $old 0000000000000000000000000000000000000000 $ref" >&3
441 freshen_loose_objects "$old"
442 echo "$old 0000000000000000000000000000000000000000 $ref"
443 done
444 LC_ALL=C join -v 2 .refs-before .refs-after |
445 while read ref new; do
446 echo "$loghhmmss 0000000000000000000000000000000000000000 $new $ref" >&3
447 freshen_loose_objects "$new"
448 echo "0000000000000000000000000000000000000000 $new $ref"
449 done
450 git for-each-ref --format='%(objectname) %(objectname) %(refname)' refs/heads
451 echo "done ref-changes %$proj% $proj"
452 } >.refs-temp 3>>"$logname"
453 if [ -S "$sockpath" ]; then
454 trap ':' PIPE
455 nc_openbsd -w 15 -U "$sockpath" <.refs-temp || :
456 trap - PIPE
458 bang config_set lastchange "$(date '+%a, %d %b %Y %T %z')"
459 bang_eval "git for-each-ref --sort=-committerdate --format='%(committerdate:iso8601)' \
460 --count=1 refs/heads >info/lastactivity"
461 ! [ -d htmlcache ] || { >htmlcache/changed; } 2>/dev/null || :
462 rm -f .delaygc .allowgc
464 [ "${cfg_autogchack:-0}" != "0" ] &&
465 [ "$(git config --get --bool girocco.autogchack 2>/dev/null)" != "false" ]
466 then
467 mv -f .refs-after .refs-last
471 # If the repository does not yet have a valid HEAD symref try to set one
472 # If an empty repository was cloned and then later becomes unempty you just
473 # lose out on the fancy "symref=HEAD:" logic and get this version instead
474 check_and_set_head || :
476 rm -f .refs-before .refs-after .refs-temp FETCH_HEAD
478 if is_banged; then
479 [ -z "$mailaddrs" ] || ! was_banged_message_sent ||
481 echo "$proj update succeeded - failure recovery"
482 echo "this status message may be disabled on the project admin page"
483 } | mailref "update@$cfg_gitweburl/$proj.git" -s "[$cfg_name] $proj update succeeded" "$mailaddrs" || :
484 bang_reset
487 progress "- [$proj] update ($(date))"