clones: set unix socket to autoflush just in case
[girocco.git] / jobd / gc.sh
blobefe455c07bc30500fb0bad76d5f7f638705e28fb
1 #!/bin/sh
3 # NOTE: additional options can be passed to git repack by specifying
4 # them after the project name, for example:
5 # gc.sh my-project -f
7 . @basedir@/shlib.sh
9 set -e
11 if [ $# -lt 1 ]; then
12 echo "Usage: gc.sh projname [extra-repack-args]" >&2
13 exit 1
16 # packing options
17 packopts="--depth=50 --window=50 --window-memory=${var_window_memory:-1g}"
19 umask 002
20 [ "$cfg_permission_control" != "Hooks" ] || umask 000
22 pidactive() {
23 if _result="$(kill -0 "$1" 2>&1)"; then
24 # process exists and we have permission to signal it
25 return 0
27 case "$_result" in *"not permitted"*)
28 # we do not have permission to signal the process
29 return 0
30 esac
31 # process does not exist
32 return 1
35 createlock() {
36 # A .lock file should only exist for much less than a second.
37 # If we see a stale lock file (> 1h old), remove it and then,
38 # just in case, wait 30 seconds for any process whose .lock
39 # we might have just removed (it's racy) to finish doing what
40 # should take much less than a second to do.
41 _stalelock="$(find "$1.lock" -maxdepth 1 -mmin +60 -print 2>/dev/null || :)"
42 if [ -n "$_stalelock" ]; then
43 rm -f "$_stalelock"
44 sleep 30
46 for _try in p p n; do
47 if (set -C; > "$1.lock") 2>/dev/null; then
48 echo "$1.lock"
49 return 0
51 # delay and try again
52 [ "$_try" != "p" ] || sleep 1
53 done
54 # cannot create lock file
55 return 1
58 # return true if there's more than one objects/pack-<sha>.pack file or
59 # ANY sha-1 files in objects
60 is_dirty() {
61 _packs=$(find objects/pack -type f -name "pack-$octet20.pack" -print | head -n 2 | wc -l)
62 if [ $_packs != 1 ] && [ $_packs != 0 ]; then
63 return 0
65 _objs=$(find objects/$octet -type f -name "$octet19" -print 2>/dev/null | head -n 1 | wc -l)
66 [ $_objs -ne 0 ]
69 # if the current directory is_gfi_mirror then repack all packs listed in gfi-packs
70 repack_gfi_packs() {
71 is_gfi_mirror || return 0
72 [ -s gfi-packs ] || return 0
73 while IFS=': ' read -r _pack _junk; do
74 if [ -s "$_pack" -a -s "${_pack%.pack}.idx" ]; then
75 git show-index < "${_pack%.pack}.idx" | cut -d ' ' -f 2
77 done < gfi-packs | \
78 git pack-objects $packopts --no-reuse-delta --delta-base-offset \
79 --non-empty $quiet objects/pack/packtmp | \
80 while read -r _newpack; do
81 rm -f objects/pack/pack-$_newpack.*
82 ln -f objects/pack/packtmp-$_newpack.pack objects/pack/pack-$_newpack.pack
83 ln -f objects/pack/packtmp-$_newpack.idx objects/pack/pack-$_newpack.idx
84 rm -f objects/pack/packtmp-$_newpack.*
85 done
86 rm -f gfi-packs
89 # HEADSHA="$(pack_is_complete /full/path/to/some.pack /full/path/to/packed-refs "$(cat HEAD)")"
90 pack_is_complete() {
91 # Must have a matching .idx file and a non-empty packed-refs file
92 [ -s "${1%.pack}.idx" ] || return 1
93 [ -s "$2" ] || return 1
94 _headsha=
95 case "$3" in
96 $octet20)
97 _headsha="$3"
99 "ref: refs/"?*|"ref:refs/"?*|"refs/"?*)
100 _headmatch="${3#ref:}"
101 _headmatch="${_headmatch# }"
102 _headmatchpat="$(echo "$_headmatch" | sed -e 's/\([.$]\)/\\\1/g')"
103 _headsha="$(grep -e "^$octet20 $_headmatchpat\$" < "$2" | \
104 cut -d ' ' -f 1)"
105 case "$_headsha" in $octet20) :;; *)
106 return 1
107 esac
110 # bad HEAD
111 return 1
112 esac
113 rm -rf pack_is_complete_test
114 mkdir pack_is_complete_test
115 mkdir pack_is_complete_test/refs
116 mkdir pack_is_complete_test/objects
117 mkdir pack_is_complete_test/objects/pack
118 echo "$_headsha" > pack_is_complete_test/HEAD
119 ln -s "$1" pack_is_complete_test/objects/pack/
120 ln -s "${1%.pack}.idx" pack_is_complete_test/objects/pack/
121 ln -s "$2" pack_is_complete_test/packed-refs
122 _count="$(git --git-dir=pack_is_complete_test rev-list --count --all 2>/dev/null || :)"
123 rm -rf pack_is_complete_test
124 [ -n "$_count" ] || return 1
125 [ "$_count" -gt 0 ] 2>/dev/null || return 1
126 echo "$_headsha"
129 proj="${1%.git}"
130 shift
131 cd "$cfg_reporoot/$proj.git"
133 trap 'e=$?; rm -f .gc_in_progress; if [ $e != 0 ]; then echo "gc failed dir: $PWD" >&2; fi' EXIT
134 trap 'exit 130' INT
135 trap 'exit 143' TERM
137 # date -R is linux-only, POSIX equivalent is '+%a, %d %b %Y %T %z'
138 datefmt='+%a, %d %b %Y %T %z'
140 if check_interval lastgc $cfg_min_gc_interval; then
141 progress "= [$proj] garbage check skip (last at $(config_get lastgc))"
142 exit 0
144 if [ -e .nogc ]; then
145 progress "x [$proj] garbage check disabled"
146 exit 0
149 # Avoid unnecessary garbage collections:
150 # 1. If lastreceive is set and is older than lastgc
151 # -AND-
152 # 2. We are not a fork (! -s alternates) -OR- lastparentgc is older than lastgc
154 # If lastgc is NOT set or lastreceive is NOT set we MUST run gc
155 # If we are a fork and lastparentgc is NOT set we MUST run gc
157 # If the repo is dirty after removing any crud we MUST run gc
159 gcstart="$(date "$datefmt")"
160 skipgc=
161 isfork=
162 [ -s objects/info/alternates ] && isfork=1
163 lastparentgcsecs=
164 [ -n "$isfork" ] && lastparentgcsecs="$(config_get_date_seconds lastparentgc || :)"
165 lastreceivesecs=
166 if lastreceivesecs="$(config_get_date_seconds lastreceive)" && \
167 lastgcsecs="$(config_get_date_seconds lastgc)" && \
168 [ $lastreceivesecs -lt $lastgcsecs ]; then
169 # We've run gc since we last received, so maybe we can skip,
170 # check if not fork or fork and lastparentgc < lastgc
171 if [ -n "$isfork" ]; then
172 if [ -n "$lastparentgcsecs" ] && \
173 [ $lastparentgcsecs -lt $lastgcsecs ]; then
174 # We've run gc since our parent ran gc so we can skip
175 skipgc=1
177 else
178 # We don't have any alternates (we're not a forK) so we can skip
179 skipgc=1
183 # be compatibile with gc.pid file from newer Git releases
185 hn="$(hostname)"
186 lockf=gc.pid
187 active=
188 if [ "$(createlock "$lockf")" ]; then
189 # If $lockf is:
190 # 1) less than 12 hours old
191 # 2) contains two fields (pid hostname) NO trailing NL
192 # 3) the hostname is different OR the pid is still alive
193 # then we exit as another active process is holding the lock
194 if [ "$(find "$lockf" -maxdepth 1 -mmin -720 -print 2>/dev/null)" ]; then
195 apid=
196 ahost=
197 read -r apid ahost ajunk < "$lockf" || :
198 if [ "$apid" ] && [ "$ahost" ]; then
199 if [ "$ahost" != "$hn" ] || pidactive "$apid"; then
200 active=1
204 else
205 echo >&2 "[$proj] unable to create gc.pid.lock file"
206 exit 1
208 if [ -n "$active" ]; then
209 rm -f "$lockf.lock"
210 echo >&2 "[$proj] gc already running on machine '$ahost' pid '$apid'"
211 exit 1
213 printf "%s %s" "$$" "$hn" > "$lockf.lock"
214 chmod 0664 "$lockf.lock"
215 mv -f "$lockf.lock" "$lockf"
217 # At this point, if .allowgc exists, it's now crud to be removed
218 rm -f .allowgc
220 # Remove any existing FETCH_HEAD
221 # There can only be a FETCH_HEAD if we've been fetching, not if we've been
222 # receiving pushes (those never create a FETCH_HEAD).
223 # And if we're fetching because we're a mirror, we know we're not fetching right
224 # now since jobd.pl never runs a project's fetch simultaneously with its gc.
225 # Therefore any existing FETCH_HEAD is junk. And it may be many megabytes if
226 # there were a lot of refs.
227 rm -f FETCH_HEAD
229 # Remove any stale pack remnants that are more than an hour old.
230 # Stale pack fragments are defined as any pack-<sha1>.ext where .ext is NOT
231 # .pack AND the corresponding .pack DOES NOT exist. A bunch of stale
232 # pack-<sha1>.idx files without their corresponding .pack files are worthless
233 # and just waste space. Normally there shouldn't be any remnants but actually
234 # this can happen when things are interrupted at just the wrong time.
235 # Note that the objects/pack directory is created by git init and should
236 # always exist.
237 find objects/pack -maxdepth 1 -type f -mmin +60 -name "pack-$octet20.?*" -print | \
238 sed -e 's/^objects\/pack\/pack-//; s/\..*$//' | LC_ALL=C sort -u | \
239 while read packsha; do
240 [ ! -e "objects/pack/pack-$packsha.pack" ] || continue
241 rm -f "objects/pack/pack-$packsha".?*
242 done
244 # Remove any stale pack .keep files that are more than 12 hours old.
245 # We don't do anything to create any permanent pack .keep files, so they must
246 # be remnants from some failed push or something. Removing the .keep will
247 # allow the pack to be properly repacked.
248 find objects/pack -maxdepth 1 -type f -mmin +720 -name "pack-$octet20.keep" -print0 | xargs -0 rm -f
250 # Remove any stale tmp_pack_* or tmp_idx_* or tmp_bitmap_* or packtmp-* files
251 # that are more than 12 hours old.
252 find objects/pack -maxdepth 1 -type f -mmin +720 -name "tmp_pack_?*" -print0 | xargs -0 rm -f
253 find objects/pack -maxdepth 1 -type f -mmin +720 -name "tmp_idx_?*" -print0 | xargs -0 rm -f
254 find objects/pack -maxdepth 1 -type f -mmin +720 -name "tmp_bitmap_?*" -print0 | xargs -0 rm -f
255 find objects/pack -maxdepth 1 -type f -mmin +720 -name "packtmp-?*" -print0 | xargs -0 rm -f
257 # Remove any stale shallow_* files that are more than 12 hours old.
258 # These can be left behind by Git >= 1.8.4.2 and < 2.0.0 when a client
259 # requests a shallow clone.
260 find . -maxdepth 1 -type f -mmin +720 -name "shallow_?*" -print0 | xargs -0 rm -f
262 # Remove any stale *.temp files in the objects area that are more than 12 hours old.
263 # This can be stale sha1.temp, or stale *.pack.temp so we kill all stale *.temp.
264 find objects -type f -mmin +720 -name "*.temp" -print0 | xargs -0 rm -f
266 # Remove any stale *.lock files in the htmlcache area that might have been left
267 # behind after an abnormal exit during an attempt to update a cached file and
268 # are more than 1 hour old.
269 ! [ -d htmlcache ] || find htmlcache -type f -mmin +60 -name "*.lock" -print0 | xargs -0 rm -f
271 # Remove any stale git-svn temp files that are more than 12 hours old.
272 # The git-svn process creates temp files with random 10 character names
273 # in the root of $GIT_DIR. Unfortunately they do not have a recognizable
274 # prefix, so we just have to kill any files with a 10-character name. We
275 # do this only for git-svn mirrors. All characters are chosen from
276 # [A-Za-z0-9_] so we can at least check that and fortunately the only
277 # collision is 'FETCH_HEAD' but that shouldn't matter.
278 # There may also be temp files with a Git_ prefix as well.
279 # We also take this opportunity to run 'git svn gc'
280 if is_svn_mirror; then
281 _randchar='[A-Za-z0-9_]'
282 _randchar2="$_randchar$_randchar"
283 _randchar4="$_randchar2$_randchar2"
284 _randchar10="$_randchar4$_randchar4$_randchar2"
285 find . -maxdepth 1 -type f -mmin +720 -name "$_randchar10" -print0 | xargs -0 rm -f
286 find . -maxdepth 1 -type f -mmin +720 -name "Git_*" -print0 | xargs -0 rm -f
287 git svn gc || :
290 # Remove any stale fast_import_crash_<pid> files that are more than 3 days old.
291 if is_gfi_mirror; then
292 find . -maxdepth 1 -type f -mmin +4320 -name "fast_import_crash_?*" -print0 | xargs -0 rm -f
295 # Skip the actual gc if .delaygc is set
296 if [ -e .delaygc ]; then
297 progress "x [$proj] garbage check delayed (except for crud removal)"
298 rm -f "$lockf"
299 exit 0
302 # Do not skip gc if the repo is dirty
303 if [ -n "$skipgc" ] && ! is_dirty; then
304 progress "= [$proj] garbage check nothing but crud removal to do (`date`)"
305 config_set lastgc "$gcstart"
306 rm -f "$lockf"
307 exit 0
310 bumptime=
311 if [ -n "$isfork" ] && [ -z "$lastparentgcsecs" ]; then
312 # set lastparentgc and then update gcstart to be at least 1 second later
313 config_set lastparentgc "$gcstart"
314 bumptime=1
316 if [ -z "$lastreceivesecs" ]; then
317 # set lastreceive and then update gcstart to be at least 1 second later
318 config_set lastreceive "$gcstart"
319 bumptime=1
321 if [ -n "$bumptime" ]; then
322 sleep 1
323 gcstart="$(date "$datefmt")"
326 progress "+ [$proj] garbage check (`date`)"
328 # safe pruning: we put all our objects in all forks, then we can
329 # safely get rid of extra ones; repacks in forks will get rid of
330 # the redundant ones again then; we carefully grab only loose
331 # objects and pack .idx and .pack files
332 forkdir="$proj"
333 if [ -d "../${forkdir##*/}" ]; then
334 # It is enough to copy objects just one level down and get_repo_list
335 # takes a regular expression (which is automatically prefixed with '^')
336 # so we can easily match forks exactly one level down from this project
337 get_repo_list "$forkdir/[^/]*:" |
338 while read fork; do
339 # Ignore forks that do not exist or are symbolic links
340 [ ! -L "$cfg_reporoot/$fork.git" -a -d "$cfg_reporoot/$fork.git" ] || \
341 continue
342 # Or do not have a non-zero length alternates file
343 [ -s "$cfg_reporoot/$fork.git/objects/info/alternates" ] || \
344 continue
345 # Match objects in parent project
346 for d in objects/??; do
347 [ "$d" != "objects/??" ] || continue
348 mkdir -p "$cfg_reporoot/$fork.git/$d"
349 ln -f "$d"/* "$cfg_reporoot/$fork.git/$d" || :
350 done
351 # Match packs in parent project
352 mkdir -p "$cfg_reporoot/$fork.git/objects/pack"
353 if [ "$(echo objects/pack/pack-*.idx)" != \
354 "objects/pack/pack-*.idx" ]; then
355 ln -f objects/pack/pack-*.pack "$cfg_reporoot/$fork.git/objects/pack" || :
356 ln -f objects/pack/pack-*.idx "$cfg_reporoot/$fork.git/objects/pack" || :
358 # Update the fork's lastparentgc date (must be current, not $gcstart)
359 GIT_DIR="$cfg_reporoot/$fork.git" git config \
360 gitweb.lastparentgc "$(date "$datefmt")"
361 done
364 quiet=; [ -n "$show_progress" ] || quiet=-q
366 git pack-refs --all
367 repack_gfi_packs
368 touch .gc_in_progress
369 rm -f .gc_failed bundles/*
370 rm -f objects/pack/pack-*.bndl
371 # We use the -A option with git repack so that unreachable objects can live
372 # on for a time as loose objects. This is particularly helpful if we just
373 # happen to be in the process of sending out a ref update for a ref that was
374 # force updated and the old ref value would have otherwise been removed by
375 # repack because it was now unreachable. Admittedly the window for gc to run
376 # and do that before we manage to send out the ref update is not large, but
377 # it would not be difficult to create such a situation. Unfortunately, when
378 # Git unpacks these unreachable objects it will give them the modification
379 # time of the *.pack file they came out of. This could be very, very old.
380 # If that happens, the subsequent git prune --expire some_time_ago will still
381 # remove the object(s) and our pending ref update will still lose out.
382 # To prevent this from happening and to get the behavior we want, we now
383 # touch the modification time of all pack-<sha>.pack files so that any
384 # loosened objects get a current time. Git does not provide any other
385 # mechanism to do this. We do not want to just touch all loose objects
386 # left after the repack because that would cause objects that were loosened
387 # previously to live on which we definitely do not want.
388 touch -c -m objects/pack/pack-$octet20.pack
389 # The git repack command may issue a 'disabling bitmap' warning for some
390 # repositories. This is perfectly normal and should be suppressed unless
391 # show_progress is set. Unfortunately that means we have to grep -v the
392 # output. And furthermore, since it's a translated message, we have to
393 # force the language to english to be sure we do it.
394 repackcmd="git repack $packopts -A -d -l $quiet $@"
395 [ -n "$show_progress" ] || \
396 repackcmd="{ LC_ALL=C $repackcmd 2>&1 || touch .gc_failed; } | grep -v 'disabling bitmap' || :"
397 eval "$repackcmd"
398 [ ! -e .gc_failed ] || exit 1
399 allpacks="$(echo objects/pack/pack-$octet20.pack)"
400 curhead="$(cat HEAD)"
401 pkrf=
402 [ ! -e packed-refs ] || pkrf=packed-refs
403 eval "reposizek=$(( $(echo 0 $(du -k $pkrf $allpacks 2>/dev/null | awk '{print $1}') | \
404 sed -e 's/ / + /g') ))"
405 # The git prune command does not take a -q or --quiet but started outputting
406 # 'Checking connectivity' progress messages in v1.7.9. However, we can
407 # suppress those by piping through cat as it only activates the progress
408 # messages when stderr is a tty. We only expire loose objects older than one
409 # day just in case there's some pending action (such as sending out a ref
410 # update) in progress that might want to examine them. This may leave us with
411 # loose objects. That's okay because at the next gc interval, we will always
412 # run gc if we see any loose objects regardless of whether or not we've seen
413 # any updates or we've received new linked objects from our parent. Note that
414 # in order to keep loose objects that just recently became unreferenced but
415 # have a very old modification date around we rely on some help from both the
416 # update.sh and hooks/pre-receive scripts. Furthermore, since Git v2.2.0
417 # (d3038d22 prune: keep objects reachable from recent objects) an unreachable
418 # object that would otherwise be pruned (because it's too old) will be kept
419 # alive by an unreachable object that refers to it that's not old enough to
420 # be pruned yet.
421 prunecmd='git prune --expire 1_day_ago'
422 [ -n "$show_progress" ] || \
423 prunecmd="{ $prunecmd 2>&1 || touch .gc_failed; } | cat"
424 eval "$prunecmd"
425 [ ! -e .gc_failed ] || exit 1
426 git update-server-info
428 # darcs:// mirrors have a xxx.log file that will grow endlessly
429 # if this is a mirror and the file exists, shorten it to 10000 lines
430 # also take this opportunity to optimize the darcs repo
431 if [ ! -e .nofetch ] && [ -n "$cfg_mirror" ]; then
432 url="$(config_get baseurl || :)"
433 case "$url" in darcs://*)
434 if [ -n "$cfg_mirror_darcs" ]; then
435 url="${url%/}"
436 basedarcs="$(basename "${url#darcs:/}")"
437 if [ -f "$basedarcs.log" ]; then
438 tail -n 10000 "$basedarcs.log" > "$basedarcs.log.$$"
439 mv -f "$basedarcs.log.$$" "$basedarcs.log"
441 if [ -d "$basedarcs.darcs" ]; then
443 cd "$basedarcs.darcs"
444 # Note that this does not optimize _darcs/inventories/ :(
445 darcs optimize
449 esac
452 # Create a matching .bndl header file for the all-in-one pack we just created
453 # but only if we're not a fork (otherwise the bundle would not be complete)
454 # and we are running at least Git version 1.7.2 (pack_is_complete always fails otherwise)
455 if [ ! -s objects/info/alternates ] && [ -n "$var_have_git_172" ]; then
456 # There should only be one pack in $allpacks but if there was a
457 # simultaneous push...
458 # The one we just created will have a .idx and will NOT have a .keep
459 pkfound=
460 pkhead=
461 for pk in $allpacks; do
462 [ -s "$pk" ] || continue
463 pkbase="${pk%.pack}"
464 [ -s "$pkbase.idx" ] || continue
465 [ ! -e "$pkbase.keep" ] || continue
466 if pkhead="$(pack_is_complete "$PWD/$pk" "$PWD/packed-refs" "$curhead")"; then
467 pkfound="$pkbase"
468 break;
470 done
471 if [ -n "$pkfound" -a -n "$pkhead" ]; then
473 echo "# v2 git bundle"
474 sed -ne "/^$octet20 refs\/[^ $tab]*\$/ p" < packed-refs
475 echo "$pkhead HEAD"
476 echo ""
477 } > "$pkbase.bndl"
478 bndletag="$("$cfg_basedir/bin/rangecgi" --etag "$pkbase.bndl" "$pkbase.pack" || :)"
479 bndlsha="$(printf '%s' "$bndletag" | git hash-object --stdin || :)"
480 if [ -n "$bndletag" ]; then
481 case "$bndlsha" in $octet20)
482 bndlshatrailer="${bndlsha#????????}"
483 bndlshaprefix="${bndlsha%$bndlshatrailer}"
484 bndlname="$(TZ=UTC date +%Y%m%d_%H%M%S)-${bndlshaprefix:-0}"
485 [ -d bundles ] || mkdir bundles
486 echo "${pkbase#objects/pack/}.bndl" > "bundles/$bndlname"
487 echo "${pkbase#objects/pack/}.pack" >> "bundles/$bndlname"
488 ln -s -f -n "$bndlname" bundles/latest
489 esac
494 # Record the size of this repo as the sum of its *.pack sizes as 1024-byte blocks
495 config_set_raw girocco.reposizek "${reposizek:-0}"
497 # We use $gcstart here to avoid a race where a push occurs during the gc itself
498 # and the next future gc could be incorrectly skipped if we used the current
499 # timestamp here instead
500 config_set lastgc "$gcstart"
501 rm -f "$lockf"
503 progress "- [$proj] garbage check (`date`)"