gc-util-functions.sh: combine loose packs as needed
[girocco.git] / jobd / gc.sh
blobf7694bd85a1cbd70b583a6204f0bc98e0b573617
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
8 . @basedir@/jobd/gc-util-functions.sh
10 set -e
12 if [ $# -lt 1 ]; then
13 echo "Usage: gc.sh projname [extra-repack-args]" >&2
14 exit 1
17 # Includes
18 _shlib_done=1
19 unset GIROCCO_SUPPRESS_AUTO_GC_UPDATE
20 . "$cfg_basedir/jobd/maintain-auto-gc-hack.sh"
21 . "$cfg_basedir/jobd/generate-auto-gc-update.sh"
22 GIROCCO_SUPPRESS_AUTO_GC_UPDATE=1 && export GIROCCO_SUPPRESS_AUTO_GC_UPDATE
24 umask 002
25 [ "$cfg_permission_control" != "Hooks" ] || umask 000
26 clean_git_env
28 pidactive() {
29 if _result="$(kill -0 "$1" 2>&1)"; then
30 # process exists and we have permission to signal it
31 return 0
33 case "$_result" in *"not permitted"*)
34 # we do not have permission to signal the process
35 return 0
36 esac
37 # process does not exist
38 return 1
41 createlock() {
42 # A .lock file should only exist for much less than a second.
43 # If we see a stale lock file (> 1h old), remove it and then,
44 # just in case, wait 30 seconds for any process whose .lock
45 # we might have just removed (it's racy) to finish doing what
46 # should take much less than a second to do.
47 _stalelock="$(find -L "$1.lock" -maxdepth 1 -mmin +60 -print 2>/dev/null)" || :
48 if [ -n "$_stalelock" ]; then
49 rm -f "$_stalelock"
50 sleep 30
52 for _try in p p n; do
53 if (set -C; >"$1.lock") 2>/dev/null; then
54 echo "$1.lock"
55 return 0
57 # delay and try again
58 [ "$_try" != "p" ] || sleep 1
59 done
60 # cannot create lock file
61 return 1
64 # The pre-receive script creates one ref log file per push but we want them to
65 # be coalesced into one ref log file per day. We are guaranteed that any files
66 # we find to coalesce are NOT currently being written to since they are always
67 # written first as temporary files and then moved into place. We attempt to
68 # transfer the most recent modification time to the coalesced log file which
69 # would step on its mod time if it were being written to directly, but if we
70 # find per-process ref log files then it must be a push project and the only
71 # thing that would write directly to the main per-day log file would be a
72 # mirror project so there's actually no conflict.
73 # Also, if the clock is wonky (or was futzed with) we may have both YYYYMMDD
74 # and YYYYMMDD.gz present in which case combine them into YYYYMMDD
75 coalesce_reflogs() {
76 [ -d reflogs ] || return 0
77 rm -f .gc_failed
78 find -L reflogs -maxdepth 1 -type f -name "[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]" -print |
79 while read -r rname; do
80 if [ -e "$rname.gz" ]; then
81 if [ -s "$rname" ]; then
82 # Presumably the .gz file must have been created before the non-gz
83 # file since it had to be uncompressed at some point therefore
84 # we need to append the non-gz contents to it but keep the non-gz
85 # contents timestamp so we rename to YYYYMMDD_ which will sort first
86 # and be picked up in the next step if we are interrupted in the middle.
87 # If a YYYYMMDD_ file already exists we append to it and transfer the
88 # timestamp. Finally we transfer the YYYYMMDD_ timestamp to the result
89 # and remove the YYYYMMDD_ temporary file leaving the result uncompressed.
90 if [ -e "${rname}_" ]; then
91 cat "$rname" >>"${rname}_"
92 touch -r "$rname" "${rname}_"
93 rm -f "$rname"
94 ! [ -e "$rname" ]
95 else
96 mv "$rname" "${rname}_"
98 gzip -d "$rname.gz" </dev/null
99 [ -e "$rname" ] && ! [ -e "$rname.gz" ]
100 cat "${rname}_" >>"$rname"
101 touch -r "${rname}_" "$rname"
102 rm -f "${rname}_"
103 else
104 # Just remove the empty file to resolve the problem
105 rm -f "$rname"
108 done
109 find -L reflogs -maxdepth 1 -type f -name "[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]_*" -print | LC_ALL=C sort |
110 while read -r rname; do
111 logname="${rname%%_*}"
112 # If someone's been futzing with the date, the file we want to
113 # append to could already have been compressed, so we just uncompress
114 # it here. The previous block guarantees we do not have both a compressed
115 # and uncompressed version present at the same time.
116 if [ -e "$logname.gz" ]; then
117 gzip -d "$logname.gz" </dev/null
118 [ -e "$logname" ] && ! [ -e "$logname.gz" ]
120 cat "$rname" >>"$logname"
121 touch -r "$rname" "$logname"
122 rm -f "$rname"
123 if [ -e "$rname" ]; then
124 >.gc_failed
125 echo "! [$proj] failed to remove $rname" >&2
126 exit 1 # will only exit subshell created by "|"
128 done
129 ! [ -e .gc_failed ]
132 # Remove any files in reflogs that are older than $cfg_reflogs_lifetime days
133 prune_reflogs() {
134 [ -d reflogs ] || return 0
135 exp="$(( ${cfg_reflogs_lifetime:-1} * 1440 ))"
136 [ $exp -gt 0 ] || exp=1440
137 [ $exp -le 43200 ] || exp=43200
138 find -L reflogs -maxdepth 1 -type f -mmin "+$exp" -exec rm -f '{}' + || :
141 # Compact any reflogs that are not today's UTC date unless a .gz version exists
142 compact_reflogs() {
143 [ -d reflogs ] || return 0
144 _td="reflogs/$(TZ=UTC date '+%Y%m%d')"
145 find -L reflogs -maxdepth 1 -type f -name "[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]" -print |
146 while read -r rname; do
147 [ "$rname" != "$_td" ] || continue
148 ! [ -e "$rname.gz" ] || continue
149 gzip -9 "$rname" </dev/null
150 done
153 # return true if there's more than one objects/pack-<sha>.pack file or
154 # ANY sha-1 files in objects or
155 # there's one pack and it's not a normal pack name or
156 # there's one pack but not any refs
157 is_dirty() {
158 _packs="$(find -L objects/pack -name "pre-auto-gc-[12].pack" -prune -o -name "*.pack" -type f -print 2>/dev/null | head -n 2)"
159 v_cnt _packscnt $_packs
160 if [ $_packscnt -gt 1 ]; then
161 return 0
163 if [ $_packscnt -eq 1 ]; then
164 # the single pack name is in $_packs
165 _packs="${_packs%.pack}"
166 _packs="${_packs#objects/pack/}"
167 case "$_packs" in
168 pack-*)
169 _packs="${_packs#pack-}"
170 if [ "${#_packs}" -lt 40 ] || [ "${_packs#*[!0-9a-fA-F]}" != "$_packs" ]; then
171 # name not exclusively 40 or more hexadecimal digits makes it dirty
172 return 0
176 # abnormal name makes it dirty
177 return 0
179 esac
181 _objs=$(find -L objects/$octet -name "$octet19*" -type f -print 2>/dev/null | head -n 1 | LC_ALL=C wc -l)
182 [ $_objs -eq 0 ] || return 0
183 [ $_packscnt -eq 1 ] || return 1
184 # we do this check last because it's potentially the most expensive;
185 # at this point we know we do not have any loose objects, but we do
186 # have one pack that's named "normally"; empty refs => dirty
187 is_empty_refs_dir
190 # combine the input pack(s) into a new pack (or possibly packs if packSizeLimit set)
191 # input pack names are read from standard input one per line delimited by the first
192 # ':', ' ' or '\n' character on the line (which allows gfi-packs to be read directly)
193 # all arguments, if any, are passed to pack-objects as additional options
194 # returns non-zero on failure AND creates .gc_failed in that case
195 combine_packs() {
196 rm -f .gc_failed
197 combine_packs_std "$@" || {
198 >.gc_failed
199 return 1
201 return 0
204 # if the current directory is_gfi_mirror then repack all packs listed in gfi-packs
205 repack_gfi_packs() {
206 [ -n "$gfi_mirror" ] || return 0
207 [ -d objects/pack ] || { rm -f gfi-packs; return 0; }
208 progress "~ [$proj] redeltifying poor quality git fast-import packs"
209 combine_packs --ignore-missing --no-reuse-delta <gfi-packs
210 rm -f gfi-packs
211 return 0
214 # pack any existing loose objects into a new _l.pack file then run prune-packed
215 # note that prune-packed is NOT run beforehand -- the caller must do that if needed
216 # loose objects need not be part of complete commits/trees as --weak-naming is used
217 pack_loose_objects() {
218 _lpacks="$(run_combine_packs </dev/null --names --loose --weak-naming --non-empty --all-progress-implied ${quiet:---progress} $packopts)"
219 if [ -n "$_lpacks" ]; then
220 # We need to identify these packs later so we don't combine_packs them
221 for _objpack in $_lpacks; do
222 rename_pack "objects/pack/pack-$_objpack" "objects/pack/pack-${_objpack}_l" || :
223 done
224 git prune-packed $quiet
228 # combine small packs into larger pack(s)
229 # we avoid any _[lo], keep, bndl or bitmap packs
230 # if the optional argument is non-empty even a single small pack will be redeltad
231 combine_small_packs() {
232 _didprogress=
233 _minsmallpacks=2
234 if [ -n "$1" ] && [ -n "$noreusedeltaopt" ]; then
235 _minsmallpacks=1
237 _lpo="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl"
238 _lpo="$_lpo --exclude-sfx _u --exclude-sfx _o --exclude-sfx _l"
239 _lpo="$_lpo --quiet --object-limit $var_redelta_threshold objects/pack"
240 while
241 _cnt="$(list_packs --count $_lpo)" || :
242 test "${_cnt:-0}" -ge $_minsmallpacks
244 [ -n "$_didprogress" ] || {
245 progress "~ [$proj] combining small packs into a single larger pack"
246 _didprogress=1
248 _newp="$(list_packs $_lpo | combine_packs --names $noreusedeltaopt)"
249 v_cnt _newc $_newp
250 # be paranoid and exit the loop if we haven't reduced the number of packs
251 [ $_newc -lt $_cnt ] || break
252 _minsmallpacks=2
253 done
254 return 0
257 # combine small _l packs into larger pack(s) using --weak-naming
258 # we avoid any non _l, keep, bndl or bitmap packs
259 # if the optional 2nd argument is non-empty even a single small pack will be redeltad
260 combine_small_loose_packs() {
261 _didprogress=
262 _minsmallpacks=2
263 if [ -n "$1" ] && [ -n "$noreusedeltaopt" ]; then
264 _minsmallpacks=1
266 _lpo="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl"
267 _lpo="$_lpo --exclude-no-sfx _l"
268 _lpo="$_lpo --quiet --object-limit $var_redelta_threshold objects/pack"
269 while
270 _cnt="$(list_packs --count $_lpo)" || :
271 test "${_cnt:-0}" -ge $_minsmallpacks
273 [ -n "$_didprogress" ] || {
274 progress "~ [$proj] combining small loose packs into a single larger pack"
275 _didprogress=1
277 _newp="$(list_packs $_lpo | combine_packs --names --weak-naming $noreusedeltaopt)"
278 # We need to identify these packs later so we don't combine_packs them
279 for _objpack in $_newp; do
280 rename_pack "objects/pack/pack-$_objpack" "objects/pack/pack-${_objpack}_l" || :
281 done
282 v_cnt _newc $_newp
283 # be paranoid and exit the loop if we haven't reduced the number of packs
284 [ $_newc -lt $_cnt ] || break
285 _minsmallpacks=2
286 done
287 return 0
290 # Unfortunately, some fetch strategies (e.g. git-svn and non-smart HTTP) lack
291 # the ability to store newly fetched objects in a pack.
292 # However, the fetch code conveniently sets .needspack just before it fetches
293 # so that it's easy to find all the loose objects that have been fetched and
294 # combine them into a pack. The --no-reuse-delta option is meaningless here
295 # since everything to be packed is a loose object and therefore not a delta so
296 # deltification will always take place.
297 make_needs_pack() {
298 [ -f .needspack ] || return 0
299 rm -f .needspackgc
300 mv -f .needspack .needspackgc
301 progress "~ [$proj] combining fetched loose objects into a pack"
302 _newp="$(find -L objects/$octet -maxdepth 1 -type f -newer .needspackgc -name "$octet19*" -print 2>/dev/null |
303 LC_ALL=C awk -F / '{print $2 $3}' |
304 run_combine_packs --objects --names $packopts --incremental --all-progress-implied $quiet --non-empty)" || {
305 # We used to fail gc here.
306 # Now, however, we just ignore the failure because we have
307 # another mechanism to handle loose objects and it's possible
308 # that the fetcher somehow brought in unconnected objects which
309 # would cause the above combine-packs to fail.
310 # By ignoring the failure and just removing the .needspack file
311 # the loose objects will be treated as "ordinary" loose objects
312 # and packed using the "--weak-naming" option which can handle
313 # broken connectivity.
314 # That's a better solution than just failing here or leaving
315 # .needspack behind to potentially continue to fail again and
316 # again.
317 _newp=
319 if [ -n "$_newp" ]; then
320 # remove the now-redundant loose objects -- this is always safe
321 # even during a concurrent push because a reprepare_packed_git
322 # will be triggered if an object that should be there is not
323 # found thereby finding it in the new pack instead
324 git prune-packed $quiet
326 rm -f .needspackgc
329 # HEADSHA="$(pack_is_complete /full/path/to/some.pack /full/path/to/packed-refs "$(cat HEAD)")"
330 pack_is_complete() {
331 # Must have a matching .idx file and a non-empty packed-refs file
332 [ -s "${1%.pack}.idx" ] || return 1
333 [ -s "$2" ] || return 1
334 _headsha=
335 case "$3" in
336 $octet20*)
337 _headsha="$3"
339 "ref: refs/"?*|"ref:refs/"?*|"refs/"?*)
340 _headmatch="${3#ref:}"
341 _headmatch="${_headmatch# }"
342 _headmatchpat="$(echo "$_headmatch" | LC_ALL=C sed -e 's/\([.$]\)/\\\1/g')"
343 _headsha="$(LC_ALL=C grep -e "^$octet20$hexdig* $_headmatchpat\$" <"$2" |
344 LC_ALL=C cut -d ' ' -f 1)"
345 case "$_headsha" in $octet20*) :;; *)
346 return 1
347 esac
350 # bad HEAD
351 return 1
352 esac
353 rm -rf pack_is_complete_test
354 mkdir pack_is_complete_test
355 mkdir pack_is_complete_test/refs
356 mkdir pack_is_complete_test/objects
357 mkdir pack_is_complete_test/objects/pack
358 echo "$_headsha" >pack_is_complete_test/HEAD
359 ln -s "$1" pack_is_complete_test/objects/pack/
360 ln -s "${1%.pack}.idx" pack_is_complete_test/objects/pack/
361 ln -s "$2" pack_is_complete_test/packed-refs
362 _count="$(git --git-dir=pack_is_complete_test rev-list --count --all 2>/dev/null)" || :
363 rm -rf pack_is_complete_test
364 [ -n "$_count" ] || return 1
365 [ "$_count" -gt 0 ] 2>/dev/null || return 1
366 echo "$_headsha"
369 # On return a "$lockf" will have been created that must be removed when gc is done
370 lock_gc() {
371 # be compatibile with gc.pid file from newer Git releases
372 lockf=gc.pid
373 hn="$(hostname)"
374 active=
375 if [ "$(createlock "$lockf")" ]; then
376 # If $lockf is:
377 # 1) less than 12 hours old
378 # 2) contains two fields (pid hostname) NO trailing NL
379 # 3) the hostname is different OR the pid is still alive
380 # then we exit as another active process is holding the lock
381 if [ "$(find -L "$lockf" -maxdepth 1 -mmin -720 -print 2>/dev/null)" ]; then
382 apid=
383 ahost=
384 read -r apid ahost ajunk <"$lockf" || :
385 if [ "$apid" ] && [ "$ahost" ]; then
386 if [ "$ahost" != "$hn" ] || pidactive "$apid"; then
387 active=1
391 else
392 echo >&2 "[$proj] unable to create gc.pid.lock file"
393 exit 1
395 if [ -n "$active" ]; then
396 rm -f "$lockf.lock"
397 echo >&2 "[$proj] gc already running on machine '$ahost' pid '$apid'"
398 exit 1
400 printf "%s %s" "$$" "$hn" >"$lockf.lock"
401 chmod 0664 "$lockf.lock"
402 mv -f "$lockf.lock" "$lockf"
405 # Create a repack subdirectory such that running repack in it will pack the
406 # same things that a pack in the normal directory would except that the pack
407 # is guaranteed to be generated in an optimized order by adding a suitable
408 # synthesized ref in the refs/tags namespace (yes, pack-objects.c really does
409 # behave differently depending on the contents of the refs/tags namespace).
410 # Before calling this, pack-refs --all MUST be performed or the wrong pack
411 # will end up being made.
413 # If a ref deletion is pushed after making the repack subdir but before the
414 # the actual repack, the discarded objects will be packed -- no big deal,
415 # they'll get discarded the next time gc runs.
417 # If a fast-forward ref update is pushed after making the repack subdir but
418 # before the actual repack, it will be picked up and the new objects packed
419 # (subject to the normal git repack race about picking such updates up).
421 # If a non-fast-forward ref update is pushed after making the repack subdir but
422 # before the actual repack, it will be picked up like a fast-forward update but
423 # the discarded objects will be included like a ref deletion (until the next
424 # scheduled gc takes place).
426 # We retain a copy of the original packed-refs file as repack/packed-refs.orig
427 # If ref deletions come in while we're repacking, the original packed-refs
428 # file will be modified, but we'll still pack the deleted ref(s).
429 # If the packed-refs.orig file is used to create the bundle header we avoid
430 # a situation where the bundle contains a ref state that never actually
431 # existed in reality (for example a new branch is pushed and then an old
432 # branch deleted afterwards -- the deletion would show up in the bundle
433 # because it will cause the original packed-refs file to be re-written, but
434 # the new branch creation will not unless we do another pack-refs which might
435 # lead to having in incomplete bundle). Therefore we want to keep a copy of
436 # the original packed-refs file around. We do the same thing for HEAD.
438 # It's possible that the "objects" subdirectory is a symbolic link.
439 # Git does support this. However, during the repacking process, new packs
440 # will be created in repack/alt/pack and then moved into objects/pack.
441 # In order for this to work seemlessly, they must both be on the same
442 # filesystem. But when objects (or even objects/pack) is a symbolic link they
443 # might not be. For this reason a "repack" subdirectory is created under
444 # objects/pack and the repack/alt/pack directory symbolicly linked to it.
446 # Git allows not just HEAD to be a symbolic-ref, but any ref anywhere in the
447 # refs namespace. We are concerned about ref name collisions and getting the
448 # right tag set to get an optimal pack. We can safely duplicate the ref space
449 # under refs/heads, refs/notes and refs/remotes without any risk of unwanted
450 # collisions and this will likely make over 99%+ of all symbolic refs found
451 # in the wild work properly. Girocco itself never creates any symbolic refs
452 # inside the refs namespace; this is a nod to simultaneously using a Girocco
453 # repository for other purposes.
454 make_repack_dir() {
455 ! [ -d repack ] || rm -rf repack
456 ! [ -d repack ] || { echo >&2 "[$proj] cannot remove repack subdirectory"; exit 1; }
457 [ -d objects/pack ] || mkdir -p objects/pack
458 ! [ -d objects/pack/repack ] || rm -rf objects/pack/repack
459 ! [ -d objects/pack/repack ] || { echo >&2 "[$proj] cannot remove objects/pack/repack subdirectory"; exit 1; }
460 mkdir repack repack/refs repack/alt objects/pack/repack
461 [ -d info ] || mkdir info
462 ln -s ../config repack/config
463 ln -s ../info repack/info
464 ln -s ../objects repack/objects
465 ln -s "$PWD/objects/pack/repack" repack/alt/pack
466 ln -s ../../refs repack/refs/refs
467 ! [ -d logs ] || ln -s ../logs repack/logs
468 ! [ -d worktrees ] || ln -s ../worktrees repack/worktrees
469 _lines=$(( $(LC_ALL=C wc -l <packed-refs) ))
470 cat HEAD >repack/HEAD.orig
471 >repack/packed-refs.extra
472 _xtralines=0
473 cat packed-refs >repack/packed-refs.orig
474 if [ $(LC_ALL=C wc -l <repack/packed-refs.orig) -ne "$_lines" ]; then
475 echo >&2 "[$proj] error: make_repack_dir failed original packed-refs line count sanity check"
476 exit 1
478 if [ "${cfg_fetch_stash_refs:-0}" = "0" ]; then
479 # migrate any refs/stash or refs/tgstash lines to repack/packed-refs.extra
480 <repack/packed-refs.orig LC_ALL=C awk -v xtra="repack/packed-refs.extra" '
481 BEGIN { peeling = 0 }
482 NR == 1 && /^#/ { print; next; }
483 peeling && /^\^/ { print >>xtra; next; }
484 /^[0-9A-Fa-f][0-9A-Fa-f][0-9A-Fa-f][0-9A-Fa-f]+ refs\/(stash|tgstash)(\/|$)/ {
485 peeling = 1
486 print >>xtra
487 next
489 { peeling = 0; print; }
490 ' >repack/packed-refs.new
491 _xtralines="$(( $(LC_ALL=C wc -l <repack/packed-refs.extra) + 0 ))"
492 _newlines="$(( $(LC_ALL=C wc -l <repack/packed-refs.new) + 0 ))"
493 if [ "$(( $_newlines + $_xtralines ))" -ne "$_lines" ]; then
494 echo >&2 "[$proj] error: make_repack_dir failed packed-refs.extra line count sanity check"
495 exit 1
497 mv -f repack/packed-refs.new repack/packed-refs.orig
498 _lines="$_newlines"
500 # Note: Git v1.5.0 introduced the "# pack-refs with:" header line for the packed-refs file
501 sed '/^# pack-refs/d; s, refs/, refs/!/,' <repack/packed-refs.orig >repack/packed-refs
502 nohead=
503 headref="$(git rev-parse --verify --quiet HEAD)" || :
504 if [ -n "$headref" ]; then
505 echo "$headref refs/!=/HEAD" >>repack/packed-refs
506 echo "$headref refs/heads/!" >>repack/packed-refs
507 nohead='\, refs/heads/!$,d; '
508 _lines=$(( $_lines + 2 ))
510 if [ $(( $(LC_ALL=C wc -l <repack/packed-refs) + 1 )) -ne "$_lines" ]; then
511 echo >&2 "[$proj] error: make_repack_dir failed packed-refs initial line count sanity check"
512 exit 1
514 sed -n "$nohead"'\, refs/heads/,p; \, refs/notes/,p; \, refs/remotes/,p' <repack/packed-refs.orig >>repack/packed-refs
515 _newlines="$(( $(LC_ALL=C wc -l <repack/packed-refs) ))"
516 if [ $(( $_newlines + 1 )) -lt "$_lines" ]; then
517 echo >&2 "[$proj] error: make_repack_dir failed packed-refs extra line count sanity check"
518 exit 1
520 _lines="$_newlines"
521 optref="$(git rev-list -n 1 --all 2>/dev/null)" || :
522 if [ -n "$optref" ]; then
523 echo "$optref refs/tags/!" >>repack/packed-refs
524 _lines=$(( $_lines + 1 ))
525 echo "$optref" >repack/HEAD
526 else
527 cat HEAD >repack/HEAD
529 if [ $(LC_ALL=C wc -l <repack/packed-refs) -ne "$_lines" ]; then
530 echo >&2 "[$proj] error: make_repack_dir failed packed-refs line count sanity check"
531 exit 1
535 # Remove any crud that's been left behind by interrupted operations
536 # that did not clean up after themselves
537 remove_crud() {
538 # Remove any existing FETCH_HEAD
539 # There can only be a FETCH_HEAD if we've been fetching, not if we've been
540 # receiving pushes (those never create a FETCH_HEAD).
541 # And if we're fetching because we're a mirror, we know we're not fetching right
542 # now since jobd.pl never runs a project's fetch simultaneously with its gc.
543 # Therefore any existing FETCH_HEAD is junk. And it may be many megabytes if
544 # there were a lot of refs.
545 rm -f FETCH_HEAD
547 # remove any existing pack_is_complete_test or repack subdirectories
548 # If either exists when this function is called it's crud
549 rm -rf pack_is_complete_test repack objects/pack/repack
551 # Remove any stale pack remnants that are more than an hour old.
552 # Stale pack fragments are defined as any pack-<sha1>.ext where .ext is NOT
553 # .pack AND the corresponding .pack DOES NOT exist. A bunch of stale
554 # pack-<sha1>.idx files without their corresponding .pack files are worthless
555 # and just waste space. Normally there shouldn't be any remnants but actually
556 # this can happen when things are interrupted at just the wrong time.
557 # Note that the objects/pack directory is created by git init and should
558 # always exist.
559 find -L objects/pack -maxdepth 1 -type f -mmin +60 -name "pack-$octet20*.?*" -print |
560 LC_ALL=C sed -e 's/^objects\/pack\/pack-//; s/\..*$//' | LC_ALL=C sort -u |
561 while read packsha; do
562 ! [ -e "objects/pack/pack-$packsha.pack" ] || continue
563 rm -f "objects/pack/pack-$packsha".?*
564 done
566 # Remove any stale tmp reflogs files that are more than one hour old.
567 # Since they are created only while the pre-receive hook is running and
568 # all it does is process a bunch of refs passed to it on standard input
569 # it's inconceivable that it would ever take as much as an hour to run.
570 if [ -d reflogs ]; then
571 find -L reflogs -maxdepth 1 -type f -mmin +60 -name "tmp_*" -exec rm -f '{}' + || :
574 # Remove any stale object tmp_obj_* files that are more than 3 hours old.
575 # Really these files should only exist very briefly so there shouldn't be any
576 # but things happen that can end up leaving them behind.
577 find -L objects/$octet -maxdepth 1 -type f -mmin +180 -name "tmp_obj_?*" -exec rm -f '{}' + 2>/dev/null || :
579 # Remove any stale pack .keep files that are more than 12 hours old.
580 # We don't do anything to create any permanent pack .keep files, so they must
581 # be remnants from some failed push or something. Removing the .keep will
582 # allow the pack to be properly repacked.
583 find -L objects/pack -maxdepth 1 -type f -mmin +720 -name "pack-$octet20*.keep" -exec rm -f '{}' + || :
585 # Remove any stale tmp_pack_*, tmp_idx_*, tmp_bitmap_*, packtmp-* or .tmp-*-pack* files
586 # that are more than 12 hours old.
587 find -L objects/pack -maxdepth 1 -type f -mmin +720 \( \
588 -name "tmp_pack_?*" -o -name "tmp_idx_?*" -o -name "tmp_bitmap_?*" -o \
589 -name "packtmp-?*" -o -name ".tmp-?*-pack*" \
590 \) -exec rm -f '{}' + || :
592 # Remove any stale incoming-* object quarantine directories that are
593 # more than 12 hours old. These are new with Git >= 2.11.0.
594 find -L objects -maxdepth 1 -type d -name 'incoming-?*' -mmin +720 \
595 -exec rm -rf '{}' + || :
597 # Remove any stale shallow_* files that are more than 12 hours old.
598 # These can be left behind by Git >= 1.8.4.2 and < 2.0.0 when a client
599 # requests a shallow clone. Also discard stale .refs-temp* and
600 # .refs-new* files at the same time.
601 find -L . -maxdepth 1 -type f -mmin +720 \( \
602 -name "shallow_?*" -o -name ".refs-temp*" -o -name ".refs-new*" \
603 \) -exec rm -f '{}' + || :
605 # Remove any stale cmbnpcks-* dirs that are more than 12 hours old.
606 # These can be left behind by abnormal exits (e.g. power failure).
607 find -L . -maxdepth 1 -type d -mmin +720 -name "cmbnpcks-?*" \
608 -exec rm -rf '{}' + || :
610 # Remove any stale *.temp files in the objects area that are more than 12 hours old.
611 # This can be stale sha1.temp, or stale *.pack.temp so we kill all stale *.temp.
612 find -L objects -type f -mmin +720 -name "*.temp" -exec rm -f '{}' + || :
614 # Remove any stale *.lock files in the htmlcache area that might have been left
615 # behind after an abnormal exit during an attempt to update a cached file and
616 # are more than 1 hour old.
617 ! [ -d htmlcache ] || find -L htmlcache -type f -mmin +60 -name "*.lock" -exec rm -f '{}' + || :
619 # Remove any stale git-svn temp files that are more than 12 hours old.
620 # The git-svn process creates temp files with random 10 character names
621 # in the root of $GIT_DIR. Unfortunately they do not have a recognizable
622 # prefix, so we just have to kill any files with a 10-character name. We
623 # do this only for git-svn mirrors. All characters are chosen from
624 # [A-Za-z0-9_] so we can at least check that and fortunately the only
625 # collision is 'FETCH_HEAD' but that shouldn't matter.
626 # There may also be temp files with a Git_ prefix as well.
627 if [ -n "$svn_mirror" ]; then
628 _randchar='[A-Za-z0-9_]'
629 _randchar2="$_randchar$_randchar"
630 _randchar4="$_randchar2$_randchar2"
631 _randchar10="$_randchar4$_randchar4$_randchar2"
632 find -L . -maxdepth 1 -type f -mmin +720 -name "$_randchar10" -exec rm -f '{}' + || :
633 find -L . -maxdepth 1 -type f -mmin +720 -name "Git_*" -exec rm -f '{}' + || :
636 # Remove any stale fast_import_crash_<pid> files that are more than 3 days old.
637 if [ -n "$gfi_mirror" ]; then
638 find -L . -maxdepth 1 -type f -mmin +4320 -name "fast_import_crash_?*" -exec rm -f '{}' + || :
641 # Remove any stale core or *.core or core.* files that are more than 3 days old.
642 find -L . -maxdepth 1 -type f -mmin +4320 \( -name "core" -o -name "*.core" -o -name "core.*" \) \
643 -exec rm -f '{}' + || :
647 ## Garbage Collection Types
649 ## There are two kinds of possible garbage collection (gc) operations:
651 ## 1. A normal, full gc
652 ## 2. A "mini" gc
654 ## If the full garbage collection interval has expired (or gc has never been
655 ## run), then a normal, full gc will take place. Otherwise, a "mini" gc will
656 ## take place if the file .needsgc exists.
658 ## A "mini" gc is similar to "git gc --auto" in that it may not end up actually
659 ## doing anything unless the right conditions are present so it's not a burden
660 ## to run it often. If the file .needsgc exists, a "mini" gc will occur at
661 ## the next opportunity.
663 ## See the docs/technical/gc.txt and docs/technical/gc-mini.txt files for more
664 ## of the gory details of how garbage collection is performed.
666 ## Note, however, that the .nogc file suppresses ALL gc activity (normal or mini).
669 proj="${1%.git}"
670 shift
671 cd "$cfg_reporoot/$proj.git"
672 [ -d objects/pack ] || { rm -f gfi-packs; mkdir -p objects/pack; }
673 mirror_url="$(get_mirror_url)" || :
674 svn_mirror=
675 ! is_svn_mirror_url "$mirror_url" || svn_mirror=1
676 gfi_mirror=
677 if [ -f gfi-packs ] && [ -s gfi-packs ] && is_gfi_mirror_url "$mirror_url"; then
678 gfi_mirror=1
681 # If git config --bool --get girocco.redelta is explicitly false then automatic
682 # redelta when there are less than $var_redelta_threshold objects will be suppressed.
683 # On the other hand, if git config --get girocco.redelta is "always" then, on a full
684 # gc only, for the final repack, deltas will always be recomputed.
685 # This can be set on a per-project basis to avoid unusual pathological gc behavior.
686 # Setting this will hurt efficiency of the affected repository.
687 # Note that fast-import packs ALWAYS get new deltas regardless of this setting.
688 noreusedeltaopt="--no-reuse-delta"
689 [ "$(git config --bool --get girocco.redelta 2>/dev/null || :)" != "false" ] || noreusedeltaopt=
690 alwaysredelta=
691 [ "$(git config --get girocco.redelta 2>/dev/null || :)" != "always" ] || alwaysredelta=1
693 # Extract any -f or -F or --no-reuse-object or --no-reuse-delta options
694 # to be compatible with the old and new gc.sh versions and avoid ugly argument
695 # duplication in process lists at the same time
696 # Any options found will override the "girocco.redelta" setting
697 recompress=
698 idx=$#
699 while [ $idx -gt 0 ]; do
700 idx=$(( $idx - 1 ))
701 opt="$1"
702 shift
703 case "$opt" in
704 -f|--no-reuse-delta)
705 alwaysredelta=1
706 continue
708 -F|--no-reuse-object)
709 alwaysredelta=1
710 recompress=1
711 continue
713 -?*)
716 printf >&2 '%s\n' "bad non-option argument: $opt"
717 echo >&2 "(Did you perhaps intend to use a --xxx=yyy form?)"
718 exit 1
719 esac
720 [ -z "$opt" ] || set -- "$@" "$opt"
721 done
722 if [ -n "$alwaysredelta" ]; then
723 noreusedeltaopt="--no-reuse-delta"
724 [ -z "$recompress" ] || noreusedeltaopt="--no-reuse-object"
727 trap 'e=$?; rm -f .gc_in_progress; if [ $e != 0 ]; then echo "gc failed dir: $PWD" >&2; fi' EXIT
728 trap 'exit 130' INT
729 trap 'exit 143' TERM
731 # date -R is linux-only, POSIX equivalent is '+%a, %d %b %Y %T %z'
732 datefmt='+%a, %d %b %Y %T %z'
734 isminigc=
735 if [ "${force_gc:-0}" = "0" ] && check_interval lastgc $cfg_min_gc_interval; then
736 if [ -e .needsgc ]; then
737 isminigc=1
738 else
739 progress "= [$proj] garbage check skip (last at $(config_get lastgc))"
740 exit 0
743 if [ -e .nogc ]; then
744 progress "x [$proj] garbage check disabled"
745 exit 0
747 if ! [ -e .nofetch ] && [ -e .clone_in_progress ] && ! [ -e .clone_failed ]; then
748 progress "x [$proj] garbage check disabled (clone in progress)"
749 exit 0
751 if [ -z "$isminigc" ] && [ -e .delaygc ] && [ -e .needsgc ]; then
752 # Eligible for a full gc but .delaygc is set so it would be skipped
753 # However .needsgc is also set so transform it into a mini instead
754 isminigc=1
755 progress "~ [$proj] garbage check delayed but checking mini because .needsgc"
758 if [ -n "$isminigc" ]; then
759 # Perform a "mini" gc
760 # Note that .delaygc is ignored here as that's only intended for full gc
761 lock_gc
762 rm -f .allowgc .needsgc
763 rm -f objects/pack/pack-*_[rful].keep
764 remove_crud
765 coalesce_reflogs
766 prune_reflogs
767 compact_reflogs
768 maintain_auto_gc_hack
769 generate_auto_gc_update
770 miniactive=
771 if [ -f .needspack ]; then
772 miniactive=1
773 progress "+ [$proj] mini garbage check ($(date))"
774 make_needs_pack
776 if [ -z "$cfg_delay_gfi_redelta" ] && [ -n "$gfi_mirror" ]; then
777 # $Girocco::Config::delay_gfi_redelta is false, force redeltification now
778 if [ -z "$miniactive" ]; then
779 miniactive=1
780 progress "+ [$proj] mini garbage check ($(date))"
782 repack_gfi_packs
784 if lotsa_loose_objects; then
785 if [ -z "$miniactive" ]; then
786 miniactive=1
787 progress "+ [$proj] mini garbage check ($(date))"
789 pack_loose_objects
791 # If there aren't at least 10 non-keep, non-bitmap, non-bndl packs then
792 # don't actually process them yet
793 lpo="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl --quiet"
794 packcnt="$(list_packs --count $lpo objects/pack)" || :
795 if [ "${packcnt:-0}" -ge 10 ]; then
796 if [ -z "$miniactive" ]; then
797 miniactive=1
798 progress "+ [$proj] mini garbage check ($(date))"
800 # if we have at least 10 packs go ahead and pack all refs now too
801 git pack-refs --all --prune
802 if [ -n "$gfi_mirror" ]; then
803 repack_gfi_packs
804 packcnt="$(list_packs --count $lpo objects/pack)" || :
806 # if repack_gfi_packs dropped the pack count to < 10 don't combine
807 if [ "${packcnt:-0}" -ge 10 ]; then
808 combine_small_packs
809 combine_small_loose_packs
810 packcnt="$(list_packs --count $lpo objects/pack)" || :
812 # if we still have more than 10 packs trigger a full gc
813 if [ "${packcnt:-0}" -ge 10 ]; then
814 # We shouldn't be in a .delaygc state at this point, but if
815 # we are then nuke it because we really need a full gc now
816 rm -f .delaygc
817 git config --unset gitweb.lastgc
818 rm -f "$lockf"
819 git update-server-info # just in case
820 progress "- [$proj] mini garbage check triggering full gc too many packs ($(date))"
821 exit 0
824 rm -f "$lockf"
825 if [ -n "$miniactive" ]; then
826 git update-server-info
827 progress "- [$proj] mini garbage check ($(date))"
828 else
829 progress "= [$proj] mini garbage check nothing but crud removal to do ($(date))"
831 exit 0
834 # Avoid unnecessary garbage collections:
835 # 1. If lastreceive is set and is older than lastgc
836 # -AND-
837 # 2. We are not a fork (is_empty_alternates_file) -OR- lastparentgc is older than lastgc
839 # If lastgc is NOT set or lastreceive is NOT set we MUST run gc
840 # If we are a fork and lastparentgc is NOT set we MUST run gc
842 # If the repo is dirty after removing any crud we MUST run gc
844 gcstart="$(date "$datefmt")"
845 skipgc=
846 isfork=
847 is_empty_alternates_file objects/info/alternates || isfork=1
848 lastparentgcsecs=
849 [ -z "$isfork" ] || lastparentgcsecs="$(config_get_date_seconds lastparentgc)" || :
850 lastreceivesecs=
851 if lastreceivesecs="$(config_get_date_seconds lastreceive)" &&
852 [ "${force_gc:-0}" = "0" ] &&
853 lastgcsecs="$(config_get_date_seconds lastgc)" &&
854 [ $lastreceivesecs -lt $lastgcsecs ]; then
855 # We've run gc since we last received, so maybe we can skip,
856 # check if not fork or fork and lastparentgc < lastgc
857 if [ -n "$isfork" ]; then
858 if [ -n "$lastparentgcsecs" ] &&
859 [ $lastparentgcsecs -lt $lastgcsecs ]; then
860 # We've run gc since our parent ran gc so we can skip
861 skipgc=1
863 else
864 # We don't have any alternates (we're not a forK) so we can skip
865 skipgc=1
869 # Prevent any other simultaneous gc operations
870 lock_gc
872 # At this point, if .allowgc or .gc_failed exists, it's now crud to be removed
873 rm -f .allowgc .gc_failed
875 # Ideally we would do this in post-receive, but that would mean duplicating the
876 # logic so it's available in the chroot jail and that's highly undesirable
877 # Instead, since the first gc will be triggered immediately following the first
878 # push, we do the check here as it's quick and harmless if HEAD is already valid
879 check_and_set_head || :
881 # Always get rid of crud
882 remove_crud
884 # Always perform reflogs maintenance
885 coalesce_reflogs
886 prune_reflogs
887 compact_reflogs
889 # Always maintain auto gc hack
890 maintain_auto_gc_hack
891 generate_auto_gc_update
893 # Run 'git svn gc' now for svn mirrors
894 if [ -n "$svn_mirror" ]; then
895 git svn gc || :
898 # Skip the actual gc if .delaygc is set
899 if [ -e .delaygc ]; then
900 progress "x [$proj] garbage check delayed (except for crud removal)"
901 rm -f "$lockf"
902 exit 0
905 # Do not skip gc if the repo is dirty
906 if [ -n "$skipgc" ] && ! is_dirty; then
907 progress "= [$proj] garbage check nothing but crud removal to do ($(date))"
908 config_set lastgc "$gcstart"
909 rm -f "$lockf"
910 exit 0
913 bumptime=
914 if [ -n "$isfork" ] && [ -z "$lastparentgcsecs" ]; then
915 # set lastparentgc and then update gcstart to be at least 1 second later
916 config_set lastparentgc "$gcstart"
917 bumptime=1
919 if [ -z "$lastreceivesecs" ]; then
920 # set lastreceive and then update gcstart to be at least 1 second later
921 config_set lastreceive "$gcstart"
922 bumptime=1
924 if [ -n "$bumptime" ]; then
925 sleep 1
926 gcstart="$(date "$datefmt")"
929 progress "+ [$proj] garbage check ($(date))"
931 newdeltas=
932 [ -z "$alwaysredelta" ] || newdeltas="$noreusedeltaopt"
933 if [ -z "$newdeltas" ] && [ -n "$gfi_mirror" ]; then
934 if [ $(list_packs --exclude-no-idx --count objects/pack) -le \
935 $(list_packs --exclude-no-idx --count --quiet --only gfi-packs) ]; then
936 # Don't bother with repack_gfi_packs since everything's being repacked
937 newdeltas="--no-reuse-delta"
940 if [ -z "$newdeltas" ] && [ -n "$noreusedeltaopt" ] &&
941 [ $(list_packs --all --exclude-no-idx --count-objects objects/pack) -le $var_redelta_threshold ]; then
942 # There aren't enough objects to worry about so just redelta to get the best pack
943 newdeltas="--no-reuse-delta"
945 if [ -z "$newdeltas" ]; then
946 # Since we're not going to recompute deltas overall, we need to do the
947 # "mini" maintenance so that we can get more optimal deltas
948 [ -z "$noreusedeltaopt" ] || make_needs_pack
949 repack_gfi_packs
950 force_single_pack_redelta=
951 [ -n "$gfi_mirror" ] || [ -n "$svn_mirror" ] || force_single_pack_redelta=1
952 [ -z "$noreusedeltaopt" ] || combine_small_packs $force_single_pack_redelta
953 [ -z "$noreusedeltaopt" ] || combine_small_loose_packs $force_single_pack_redelta
957 ## Safe Pruning In Forks
959 ## We are about to perform garbage collection. We do NOT use the "git gc" or
960 ## the "git repack" commands directly as they do not provide enough control over
961 ## the fine details. However, we DO maintain a "gc.pid" file during our garbage
962 ## collection so that a simultaneous "git gc" by an administrator will be
963 ## blocked (and similarly we refuse to start garbage collection if we cannot
964 ## create the "gc.pid" file).
966 ## When we say "gc" in the below description we are referring to our "gc.sh"
967 ## script, NOT the "git gc" command.
969 ## If the project we are running garbage collection (gc) on has any forks we
970 ## must be careful not to remove any objects that while no longer referenced by
971 ## this project (the parent) are still referenced by one or more forks (the
972 ## children) otherwise the children will become corrupt and we can't abide
973 ## corrupt children.
975 ## One way to accomplish this is to simply hard-link all currently existing
976 ## loose objects and packs in the parent into all the children that refer to the
977 ## parent (via a line in their objects/info/alternates file) before beginning
978 ## the gc operation and then relying on a subsequent gc in the child to clean up
979 ## any excess objects/packs. We used to use this strategy but it's very
980 ## inefficient because:
982 ## 1. The disk space used by the old pack(s)/object(s) will not be reclaimed
983 ## until all children (and their children, if any) run gc by which time
984 ## it's quite possible the topmost parent will have run gc again and
985 ## hard-linked yet another old pack down to its children (not to mention
986 ## loose objects).
988 ## 2. When using the "-A" option with "git repack", any new objects in the
989 ## parent that are not referenced by children will continually get
990 ## exploded out of the hard-linked pack in the children whenever the
991 ## children run gc.
993 ## 3. To avoid suboptimal and/or unnecessarily many packs being hard-linked
994 ## into child forks, we must run the "mini" gc maintenance before we
995 ## perform the hard-linking into the children which provides yet another
996 ## source of inefficiency.
998 ## While we were still using the "-A" option to "git repack" (that was not
999 ## always the case) to guarantee we can access old ref values for long enough
1000 ## to send out a meaningful mail.sh notification, another, more efficient,
1001 ## option became available to prevent corruption of child forks that continue
1002 ## to refer to objects that are no longer reachable from any ref in the parent.
1004 ## The only things that need be copied (or hard-linked) into the child fork(s)
1005 ## are those objects that have become unreachable from any ref in the parent.
1007 ## When we were using the "git repack -A -d" + "git prune --expire=1.day.ago"
1008 ## technique, the only objects that could ever be removed were loose objects
1009 ## that "git prune" determined were expired. In that case, loose objects were
1010 ## all that need be hard-linked down to child forks in order to avoid
1011 ## corruption of any child fork(s).
1013 ## The "git repack -A -d" + "git prune --expire=1.day.ago" + hard-linking loose
1014 ## objects to child forks technique remains fundamentally sound from the
1015 ## perspective of supporting simultaneous gc and push and keeping newly
1016 ## unreachable objects around long enough to be sure we can send out meaningful
1017 ## ref change notifications and never corrupting any child forks and never
1018 ## persisting the lifetime of large old packs containing mostly duplicate or
1019 ## unreachable objects as gc percolates through a project's entire fork tree.
1021 ## However, that technique suffers from one potential prodigious pitfall.
1023 ## Unreachable objects come flying out of their packs to splatter all over the
1024 ## objects subdirectories possibly creating a huge, inefficient mess.
1026 ## Often this is not an issue. Even with a lot of rebasing going on, usually
1027 ## the only objects that will splatter are some commits, trees and the odd blob
1028 ## here and there. Not enough to be overly concerned about.
1030 ## However, for the reppository that frequently experiences a lot of non-fast-
1031 ## forward updates and/or outright ref deletion, the number of objects suddenly
1032 ## popping out of their packs at "git repack -A -d" time can be overwhelming.
1034 ## To avoid this issue we now use a four phase pack creation strategy.
1035 ## This will result in creation of up to four packs (instead of at most one).
1037 ## I. A complete pack (with bitmaps if appropriate) gets created including
1038 ## only "reachable" objects from all refs/... refs plus HEAD. This will
1039 ## also serve as the virtual bundle for the repository.
1041 ## II. A pack of recently-became-unreachable objects and friends is created.
1042 ## (The "friends" are ref logs, linked working tree HEADs and indicies.)
1043 ## Because both the pre-receive and update.sh script record all ref
1044 ## changes we can easily choose the cut off point for "recently".
1045 ## It is only the fact we maintain those logs in the reflogs subdirectory
1046 ## that allows this step to be possible.
1048 ## III. If the repository has any forks with a non-zero length alternates file,
1049 ## yet another pack of "--keep-unreachable" objects is generated that will
1050 ## not actually be kept in the parent, but hard-linked into all the forks.
1052 ## IV. Finally, after running "git prune-packed", any remaining loose objects
1053 ## are migrated into a pack of their own.
1055 ## We then remove any non-.keep packs that existed before we started the
1056 ## process being careful to keep any same-pack pushes for the "Push Pack Redux"
1057 ## race condition (see docs/technical/gc.txt).
1059 ## By using "git pack-objects" directly we are able to accomplish this with
1060 ## very little additional effort.
1062 ## The packs produced by (III) are treated almost like ".keep" packs by child
1063 ## forks in that the objects in them are never repacked into any other
1064 ## "--keep-unreachable" packs (but they can migrate into phase I or II packs)
1065 ## and those phase III packs are then hard-linked into any grandchild forks.
1067 ## This avoids the space explosion that could occur if each fork level ended
1068 ## up duplicating the "--keep-unreachable" pack space by repacking those
1069 ## objects (essentially breaking the hard-link to the single copy of those
1070 ## objects).
1072 ## While it is true that each level of forks could potentially add yet another
1073 ## phase III pack to be hard-linked down to its children, such packs will only
1074 ## include unreachable objects not already in any phase III packs that were
1075 ## received from the parent.
1077 ## The space for the phase III packs will not be reclaimed until the gc
1078 ## finishes percolating through the entire "fork tree" of a project.
1080 ## This is not much different than the "git repack -A -d" situation where
1081 ## all the loose objects are hard-linked down into child forks. In that
1082 ## case forks that actually need any of those objects could gradually reduce
1083 ## the number of objects hard-linked into deeper fork levels.
1085 ## The difference with a phase III "--keep-unreachable" pack is that there
1086 ## cannot be any gradual reduction like that since it would require repacking
1087 ## the pack and breaking the hard-link thereby increasing storage space. The
1088 ## storage will instead always be reclaimed all at once when all of the
1089 ## projects in the "fork tree" complete their gc.
1091 ## However, the belief is that the huge space win by having all the
1092 ## unreachable objects packed up together far eclipses (when many objects are
1093 ## involved, the single-pack version can end up using 1/20th or less of the
1094 ## disk space compared to having them all as loose objects) any brief minor
1095 ## space savings that might occur under the "git repack -A -d" loose object
1096 ## system prior to the gc collection completing for all the projects in the
1097 ## "fork tree".
1101 ## utility functions
1104 make_packs_ugw() {
1105 find -L "$1" -maxdepth 1 -type f ! -perm -ug+w \
1106 -name "pack-$octet20*.pack" -exec chmod ug+w '{}' + || :
1107 } 2>/dev/null
1109 get_index_tree() {
1110 if [ -s "$1" ]; then
1111 GIT_INDEX_FILE="$1"
1112 export GIT_INDEX_FILE
1113 git write-tree 2>/dev/null || :
1114 unset GIT_INDEX_FILE
1118 get_detached_head() {
1119 if [ -s "$1" ] && read -r _head <"$1" 2>/dev/null; then
1120 case "$_head" in $octet20*)
1121 echo "$_head"
1122 esac
1126 # single argument must be a top-level GIT_DIR
1127 # output will be (one per line, zero or more lines) full hash
1128 # values (i.e. 40 or more hex digits) from possible detached head
1129 # "single level" refs (e.g. "FETCH_HEAD" "MERGE_HEAD" etc.) from
1130 # files in the specified directory that have a "suitable" name and
1131 # have a length >= 40 and <= 1000 where EVERY line in a file MUST
1132 # match a 40 digit (or longer) hex string or that file will be ignored.
1133 # In a nod to Git each line first has any tab and anything following truncated.
1134 # NO sorting NOR "uniq"ing NOR "--batch-check"ing is performed on the output.
1135 # Also note that "HEAD" is NOT EXCLUDED and if detached will be output!
1136 get_detached_friends() {
1137 if [ -n "$1" ] && [ -d "$1" ]; then
1138 LC_ALL=C find -H "$1" -maxdepth 1 -name '[A-Za-z]*[A-Za-z0-9]' \
1139 -type f -size +39c -size -1001c -exec awk '
1140 BEGIN {exit}
1141 function process(fn, fnt, hashes, hcnt, fl, hi) {
1142 fnt = fn; sub(/^.*\//, "", fnt)
1143 if (length(fnt) > 31 || fnt !~ /^[A-Za-z][A-Za-z0-9_-]*[A-Za-z0-9]$/) return;
1144 hcnt = 0;
1145 while (getline fl < fn) {
1146 sub(/\t.*$/, "", fl)
1147 if (length(fl) < 40 || fl !~ /^[0-9a-fA-F][0-9a-fA-F]*$/) {hcnt = 0; break;}
1148 hashes[++hcnt] = fl;
1150 close(fn)
1151 for (hi=1;hi<=hcnt;++hi) print tolower(hashes[hi]);
1153 END {for (idx=1;idx<ARGC;++idx) process(ARGV[idx])}
1154 ' '{}' '+' 2>/dev/null || :
1158 # get_worktrees_friends
1159 # single argument is a (possibly relative) path to a "worktrees" subdir
1160 # if omitted it defaults to "worktrees"
1161 # any "friends" found in there are output to stdout
1162 get_worktrees_friends() {
1163 if [ -d "${1:-worktrees}" ]; then
1164 find -L "${1:-worktrees}" -mindepth 2 -maxdepth 2 -name HEAD -type f -print |
1165 while read -r _lwth; do
1166 get_detached_head "$_lwth"
1167 get_detached_friends "${_lwth%HEAD}"
1168 get_index_tree "${_lwth%HEAD}index"
1169 done
1173 # compute_extra_reachables
1174 # create lines suitable for a packed-refs file mentioning all the
1175 # other refs we might like to keep.
1176 # the current directory MUST be set to the repository's --git-dir
1177 # the following are included:
1178 # * refs mentioned in repack/packed-refs.extra (if it exists)
1179 # * refs mentioned in reflogs/... files
1180 # * tree(s) created from index file(s)
1181 # * detached linked working tree heads
1182 # Resulting objects are tested for existence and uniqified then output
1183 # one per line under a refs/z* namespace
1184 compute_extra_reachables() {
1186 if [ -s repack/packed-refs.extra ]; then
1187 LC_ALL=C sed <repack/packed-refs.extra -n \
1188 -e 's/^\([0-9A-Fa-f][0-9A-Fa-f]*\).*$/\1/p' \
1189 -e 's/^\^\([0-9A-Fa-f][0-9A-Fa-f]*\).*$/\1/p'
1191 digits8='[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]'
1192 find -L reflogs -mindepth 1 -maxdepth 1 -type f -name "$digits8*" -exec gzip -c -d -f '{}' + |
1193 LC_ALL=C awk '{print $2; print $3}'
1194 get_detached_friends .
1195 ! [ -f index ] || get_index_tree index
1196 get_worktrees_friends
1198 is_git_dir private &&
1199 [ "$(cd objects && pwd -P)" = "$(cd private/objects && pwd -P)" ] &&
1200 [ "$(cd refs && pwd -P)" != "$(cd private/refs && pwd -P)" ]
1201 then
1202 git --git-dir=private show-ref --head --hash 2>/dev/null || :
1203 get_detached_friends private
1204 ! [ -f private/index ] || get_index_tree private/index
1205 get_worktrees_friends private/worktrees
1207 } | LC_ALL=C sort -u |
1208 git cat-file ${var_have_git_260:+--buffer} --batch-check"${var_have_git_185:+=%(objectname)}" |
1209 LC_ALL=C awk '!/missing/ {num++; print $1 " " "refs/" substr("zzzzzzzzzzzz", 1, length(num)) "/" num}'
1213 ## main gc logic
1216 # Everything else is more efficient if we do this first
1217 # The "--prune" option is the default since v1.5.0 but it serves as "documentation" here
1218 git pack-refs --all --prune
1219 [ -e packed-refs ] || >>packed-refs # should never happen...
1221 # If we have a logs directory or a worktrees directory expire the ref logs now
1222 # Note that Git itself does not use either --rewrite or --updateref, so neither do we
1223 ! [ -d logs ] && ! [ -d worktrees ] || eval git reflog expire --all "${quiet:+>/dev/null 2>&1}" || :
1225 make_repack_dir
1226 ! [ -e .gc_failed ] || exit 1
1227 rm -f .gc_in_progress # make sure
1228 touch .gc_in_progress # it's truly fresh
1229 rm -f bundles/* objects/pack/pack-*.bndl
1230 # These only exist for a brief time before the packs loose their _f suffix
1231 # "Push Pack Redux" does not apply to these since they were only ever present with _f
1232 rm -f objects/pack/pack-*_f.keep
1233 # This is perhaps a bit aggressive in that if we're suffering from "Push Pack Redux"
1234 # and somehow we get run again immediately after the run where "Push Pack Redux" happened
1235 # and we have garbage collection forced, there's just the barest, almost negligible,
1236 # possibility that the "Push Pack Redux" ref updates _still_ have not happened and we
1237 # should not be removing _r .keep files. None of the normal Girocco processing can
1238 # cause this. The second run of this script would have to use the force gc option
1239 # for it to even be possible in the first place. What's much more likely is that
1240 # the initial run of this script was somehow interrupted in the middle before it
1241 # could get rid of the _r .keep file itself in which case it's better to get rid of
1242 # it now to avoid keeping something around that would perturb our nice and neat gc
1243 rm -f objects/pack/pack-*_r.keep
1244 # We will add .keep files for _u and _l packs if and when we run phase III
1245 # Otherwise they need to not have any .keep files during phases I and II
1246 rm -f objects/pack/pack-*_[ul].keep
1248 # We need to make sure that any non-Girocco (barely tolerated) Git object creation
1249 # activity will be able to "freshen" the pack containing a pre-existing object
1250 # that's being written. This really should not be necessary as the pre-receive
1251 # hook should make sure this takes place for any incoming pushes.
1252 # However, do it here anyway just in case.
1253 make_packs_ugw objects/pack
1255 # This is only effective with Git v2.3.5 and later and it will only matter when
1256 # we are using one of the "internal_rev_list" modes of pack-objects
1257 # (the combine-packs.sh script never uses any of those modes)
1258 # The "git repack" and "git prune" commands always set this internally themselves
1259 # It makes no difference if there's no repository corruption
1260 GIT_REF_PARANOIA=1 && export GIT_REF_PARANOIA
1262 # All of the options we might want to use with pack-objects were supported
1263 # at some point prior to Git version v1.6.6 which is the minimum version that
1264 # Girocco now requires. Except for one (--use-bitmap-index). Several of them
1265 # are "boiler plate" options we always want to use so we bundle them up here.
1266 pkopt="--delta-base-offset --keep-true-parents --non-empty --all-progress-implied"
1267 # We want to use --include-tag, but before Git v2.10.1 it would leave out
1268 # "middle" tags (e.g. a tag of a tag of a commit would omit the tagged tag)
1269 # See http://repo.or.cz/git.git/b773ddea2cd3b08c for details
1270 # ("pack-objects: walk tag chains for --include-tag", 2016-09-07, v2.10.1)
1271 # This is not a free check as it matches all refs against refs/tags/ then
1272 # peels all the annotated tags and checks for inclusion. The situation in
1273 # which it would add a tag that was not already included by a reachability
1274 # trace that included tag starting points can only occur if a new tag gets
1275 # pushed during gc pointing to something that would have been packed anyway.
1276 # But, it could happen and, really, compared to gc as a whole it's not that
1277 # expensive to perform (provided we do not get an unconnected pack).
1278 [ -z "$var_have_git_2101" ] || pkopt="$pkopt --include-tag"
1279 pkopt="$pkopt ${quiet:---progress} $packopts"
1281 # The git pack-objects command only supports bitmaps if all objects are being
1282 # packed (the "--all" option) and the "--stdout" option is NOT being used.
1283 # Additionally, while packing, if any encountered reachable objects are
1284 # determined to be "not wanted" then no bitmap index will be written anyway.
1285 # While it is theoretically possible that a project with a non-empty alternates
1286 # file ends up packing all objects (because it does not actually use any of the
1287 # objects found in the alternates), it's very unlikely. And, in the unlikely
1288 # event that did occur, clients would see a message about only using one bitmap
1289 # because Git can only use one bitmap at a time and at least one of the
1290 # alternates is bound to have a bitmap. Therefore if we see a non-empty
1291 # alternates file, we disable writing bitmaps which avoids the warning and any
1292 # possibility of a client warning as well. Also if we are running anything
1293 # before Git v2.1.0 (the effective version for repack.writeBitmaps=true) then
1294 # we also always disable bitmap writing.
1295 wbmopt=
1296 [ -z "$var_have_git_210" ] || wbmopt="--write-bitmap-index"
1297 # More recent versions of pack-objects have optimizations when not using the
1298 # --local option. If we do not have any alternates it's a pointless option.
1299 # If we do have alternates we need to skip writing a bitmap and we cannot
1300 # have a bundle since it must contain all objects.
1301 if [ -n "$isfork" ]; then
1302 lclopt="--local"
1303 wbmopt=
1304 makebndl=
1305 else
1306 lclopt=
1307 makebndl=1
1311 ## Phase I
1314 wbmstr=
1315 [ -n "$wbmopt" ] || wbmstr=" (bitmaps disabled)"
1316 progress "~ [$proj] running primary full gc pack-objects$wbmstr ($(date))"
1318 gotforks=
1319 ! has_forks_with_alternates "$proj" || gotforks=1
1321 # To avoid "Push Pack Redux" (see docs/technical/gc.txt), after collecting the
1322 # initial preexisting non-keep pack list, we rename them so that an incoming push
1323 # pack cannot possibly experience a pack name collision. Git does not require
1324 # use of the "default" pack names, simply that the proper extensions are used.
1325 # We rename to insert an "_r" just before the extension to avoid "Push Pack Redux"
1326 # name collisions. Later on we may create an "unreachable" pack for hard-linking
1327 # down into forks and it will have an "_u" inserted just before its extension.
1328 packlist="$(list_packs -C objects/pack --all --exclude-no-idx --exclude-keep --quiet .)" || :
1329 oldpacks=
1330 for oldpack in $packlist; do
1331 oldpack="${oldpack%.pack}"
1332 [ -f "objects/pack/$oldpack.pack" ] || {
1333 echo >&2 "[$proj] unable to list old pack files"
1334 exit 1
1336 case "$oldpack" in pre-auto-gc-[12])
1337 # we never disturb pre-auto-gc-1 or pre-auto-gc-2 packs
1338 continue
1339 esac
1340 oldpackhex="${oldpack#pack-}"
1341 if [ "${oldpackhex#*[!0-9a-fA-F]}" != "$oldpackhex" ]; then
1342 # names not exclusively hexadecimal do not need renaming
1343 case "$oldpack" in
1344 pack-$octet20*_l)
1345 # _l packs are treated like still-unpacked loose objects
1346 continue;;
1347 *_f)
1348 # _f packs can only be left over from a previously interrupted gc;
1349 # they need to be renamed to _r now so they're not confused with
1350 # any freshly generated "final" packs (and we already removed
1351 # any pre-existing *_f.keep files so we're good to go)
1354 oldpacks="${oldpacks:+$oldpacks }$oldpack"
1355 continue;;
1356 esac
1358 rename_pack "objects/pack/$oldpack" "objects/pack/${oldpack%_f}_r" || {
1359 echo >&2 "[$proj] unable to rename old pack files"
1360 exit 1
1362 # If the oldpack has a .keep now it means a "Push Pack Redux" is actually
1363 # in progress at this moment and we need to .keep the renamed pack,
1364 # otherwise no "Push Pack Redux" has started yet or it has already finished.
1365 # In either case we're okay because if it's just finished then all ref
1366 # changes have already been made so we don't need a .keep and we will
1367 # see the ref changes and grab all the objects via a reachability trace.
1368 # If it hasn't started yet that's okay because we're done moving that
1369 # name so a complete pack will appear under the old name that we'll
1370 # leave alone.
1371 if [ -f "objects/pack/$oldpack.keep" ]; then
1372 echo "Push Pack Redux" >"objects/pack/${oldpack%_f}_r.keep"
1373 else
1374 oldpacks="${oldpacks:+$oldpacks }${oldpack%_f}_r"
1376 done
1378 # We wish to keep deltas from our last full pack so if we're not redeltaing
1379 # then make sure the .pack associated with the .bitmap has a newer mod time
1380 # (If there is no .bitmap then touch the pack with the most objects instead.)
1381 if [ -z "$newdeltas" ]; then
1382 bmpack="$(list_packs --exclude-no-bitmap --exclude-no-idx --max-matches 1 objects/pack)"
1383 [ -n "$bmpack" ] || bmpack="$(list_packs --exclude-no-idx --max-matches 1 --object-limit -1 --include-boundary objects/pack)"
1384 if [ -n "$bmpack" ] && [ -f "$bmpack" ] && [ -s "$bmpack" ]; then
1385 sleep 1
1386 touch -c "$bmpack" 2>/dev/null || :
1387 # We must touch .gc_in_progress here to avoid $bmpack looking
1388 # like it's been "freshened" when redundant packs are removed
1389 # It's okay if they have the same mod time, but POSIX does not
1390 # guarantee an ordering for the "touching" that occurs which is
1391 # why this must be a separate command but needs no "sleep 1"
1392 touch .gc_in_progress
1396 # Now we need to make sure that any "freshening" that takes place will actually
1397 # result in a "newer" modification time than the .gc_in_progress file now has
1398 sleep 1
1400 # We run git pack-objects from the repack subdirectory so we can force
1401 # optimized packs to be generated even for repositories that do not have any
1402 # tagged commits
1403 packs="$(git --git-dir=repack pack-objects </dev/null \
1404 $pkopt --all $newdeltas $lclopt ${wbmopt:---honor-pack-keep} "$@" repack/alt/pack/pack)"
1405 v_cnt packcnt $packs
1406 [ $packcnt -eq 1 ] || makebndl=
1409 ## Phase II
1412 progress "~ [$proj] running supplementary gc pack-objects ($(date))"
1414 # Add the "supplementary" refs
1415 compute_extra_reachables >>repack/packed-refs
1417 # Subtract the primary refs
1418 GIT_ALTERNATE_OBJECT_DIRECTORIES="$PWD/repack/alt"
1419 export GIT_ALTERNATE_OBJECT_DIRECTORIES
1421 # For this one we MUST use --local and MUST NOT use --write-bitmap-index
1422 # However, if there is a "logs" subdirectory we need to use --reflog
1423 # We do add it, just in case, if the linked working trees dir is present
1424 # We do not add --indexed-objects as that requires v2.2.0 and it's unclear
1425 # if it properly includes linked working tree index files or not. The
1426 # above compute_extra_reachables has already included all index trees (thereby
1427 # providing proper --indexed-objects support for all Git versions) making the
1428 # option completely unnecessary.
1429 rflopt=
1430 ! [ -d logs ] && ! [ -d worktrees ] || rflopt=--reflog
1431 spacks="$(git --git-dir=repack pack-objects </dev/null \
1432 $pkopt --honor-pack-keep --all $rflopt $newdeltas --local "$@" repack/alt/pack/pack)"
1435 ## Phase III
1438 # There's nothing to do for Phase III unless we have forks that refer to our
1439 # project from their alternates file
1440 hlpacks=
1441 upacks=
1442 if [ -n "$gotforks" ]; then
1444 progress "~ [$proj] running keep-unreachable gc pack-objects for forks ($(date))"
1446 # If we are a fork, any pre-existing _u packs need to have a .keep
1447 # for this phase and be added to the hlpacks list otherwise (we are
1448 # not a fork) pre-existing _u packs are anomalies to be treated like
1449 # regular non-_u packs
1450 if [ -n "$isfork" ]; then
1451 for upack in $(find -L objects/pack -mindepth 1 -maxdepth 1 -name "pack-$octet20*_[ul].pack" -print); do
1452 upack="${upack%.pack}"
1453 [ -e "$upack.keep" ] || echo "unreachable" >"$upack.keep"
1454 case "$upack" in *_l);;*)
1455 hlpacks="${hlpacks:+$hlpacks }${upack#objects/pack/pack-}"
1456 esac
1457 done
1459 # Using either --no-reuse-delta or --no-reuse-object together with the
1460 # --keep-unreachable option is a very, very, very bad idea when good
1461 # packs are the desired outcome. If newdeltas are being generated
1462 # then we pack to a temp name, and use combine-packs.sh to get a better
1463 # pack as the result to avoid making a bad --keep-unreachable pack
1464 pfx=
1465 [ -z "$newdeltas" ] || pfx="ku"
1466 upacks="$(git --git-dir=repack pack-objects </dev/null \
1467 $pkopt --honor-pack-keep --all $rflopt --keep-unreachable --local "$@" repack/alt/pack/${pfx}pack)"
1468 if [ -n "$upacks" ] && [ -n "$newdeltas" ]; then
1469 progress "~ [$proj] rebuilding keep-unreachable pack deltas"
1470 oldupacks="$upacks"
1471 upacks="$(
1472 printf "repack/alt/pack/${pfx}pack-%s.pack\n" $oldupacks |
1473 run_combine_packs --names --weak-naming --non-empty --all-progress-implied ${quiet:---progress} \
1474 $packopts $newdeltas "$@" repack/alt/pack/pack)"
1475 eval rm -f "$(printf \""repack/alt/pack/${pfx}pack-%s.*"\"" " $oldupacks)"
1477 for upack in $upacks; do
1478 rename_pack "repack/alt/pack/pack-$upack" "repack/alt/pack/pack-${upack}_u"
1479 done
1480 rm -f objects/pack/pack-*_[ul].keep
1481 [ -z "$hlpacks" ] && [ -z "$upacks" ] ||
1482 progress "~ [$proj] hard-linking keep-unreachable pack(s) into immediate child forks"
1484 # We have to update the lastparentgc time in the child forks even if they do not get any
1485 # new "unreachable packs" because they need to run gc just in case the parent now has some
1486 # objects that used to only be in the child so they can be removed from the child.
1487 # For example, a "patch" might be developed first in a fork and then later accepted into
1488 # the parent in which case the objects making up the patch in the child fork are now
1489 # redundant (since they're now in the parent as well) and need to be removed from the
1490 # child fork which can only happen if the child fork runs gc.
1491 lastparentgc="$(date "$datefmt")"
1493 # It is enough to copy objects just one level down and get_repo_list
1494 # takes a regular expression (which is automatically prefixed with '^')
1495 # so we can easily match forks exactly one level down from this project
1496 forkdir="$proj"
1497 get_repo_list "$forkdir/[^/:][^/:]*:" |
1498 while read fork; do
1499 # Ignore forks that do not exist or are symbolic links
1500 ! [ -L "$cfg_reporoot/$fork.git" ] && [ -d "$cfg_reporoot/$fork.git" ] ||
1501 continue
1502 # Or have an empty alternates file
1503 ! is_empty_alternates_file "$cfg_reporoot/$fork.git/objects/info/alternates" ||
1504 continue
1505 runupdate=
1506 # Match hlpacks in parent project if any
1507 if [ -n "$hlpacks" ]; then
1508 mkdir -p "$cfg_reporoot/$fork.git/objects/pack"
1509 eval ln -f "$(printf '"objects/pack/pack-%s.pack" ' $hlpacks)" \
1510 "$(printf '"objects/pack/pack-%s.idx" ' $hlpacks)" \
1511 '"$cfg_reporoot/$fork.git/objects/pack/"'
1512 runupdate=1
1514 # Match upacks in repack/alt area if any
1515 if [ -n "$upacks" ]; then
1516 mkdir -p "$cfg_reporoot/$fork.git/objects/pack"
1517 eval ln -f "$(printf '"repack/alt/pack/pack-%s_u.pack" ' $upacks)" \
1518 "$(printf '"repack/alt/pack/pack-%s_u.idx" ' $upacks)" \
1519 '"$cfg_reporoot/$fork.git/objects/pack/"'
1520 runupdate=1
1522 if ! [ -e "$cfg_reporoot/$fork.git/.needsgc" ]; then
1523 # Trigger a mini gc in the fork if it now has too many packs
1524 packs="$(list_packs --quiet --count --exclude-no-idx --exclude-keep "$cfg_reporoot/$fork.git/objects/pack")" || :
1525 if [ -n "$packs" ] && [ "$packs" -ge 20 ]; then
1526 >"$cfg_reporoot/$fork.git/.needsgc"
1529 [ -z "$runupdate" ] || git --git-dir="$cfg_reporoot/$fork.git" update-server-info
1530 # Update the fork's lastparentgc date (must be more recent than $gcstart)
1531 git --git-dir="$cfg_reporoot/$fork.git" config gitweb.lastparentgc "$lastparentgc"
1532 done
1535 # Now move any primary/supplementary packs back into objects/pack
1536 # then drop any "unfreshened" redundant packs and clear repack/alt
1538 # First make sure the primary pack(s) have the most recent mod time
1539 if [ -n "$packs" ]; then
1540 [ -z "$spacks" ] || sleep 1
1541 printf 'repack/alt/pack/pack-%s.pack\n' $packs | xargs touch -c 2>/dev/null || :
1544 # Move the packs into place but with a _f suffix and a .keep file for now
1545 for pack in $packs $spacks; do
1546 rename_pack "repack/alt/pack/pack-$pack" "objects/pack/pack-${pack}_f"
1547 [ -e "objects/pack/pack-${pack}_f.keep" ] ||
1548 echo "final" >"objects/pack/pack-${pack}_f.keep"
1549 done
1551 # It's possible that one of the $oldpacks had a .bitmap, got renamed (along
1552 # with its .bitmap) and then got "freshened" causing us to not remove it
1553 # However, if $wbmopt is set we most likely now have TWO .bitmap packs!
1554 # This can produce ugly warnings we don't want and possibly get the wrong
1555 # bitmap used since only one .bitmap file can ever be used by Git.
1556 # If this has happened, the .bitmap we want to discard will always have
1557 # an _r suffix so we can just zap any such now since it will leave the pack.
1558 [ -z "$wbmopt" ] || rm -f objects/pack/pack-*_r.bitmap || :
1560 # Remove the redundant packs that have not since been "freshened"
1561 # This does not completely eliminate the race condition window (Girocco's own
1562 # activites -- gc/fetch/receive are immune to the race) but it substantially
1563 # shrinks it down to just the time after the find but before the following rm
1564 >repack/oldpacks
1565 [ -z "$oldpacks" ] ||
1566 printf 'objects/pack/%s.pack\n' $oldpacks |
1567 LC_ALL=C sort >repack/oldpacks
1568 find -L objects/pack -maxdepth 1 -type f -name "pack-$octet20*.pack" -newer .gc_in_progress -print |
1569 LC_ALL=C sort >repack/freshened
1570 deadpacks="$(LC_ALL=C join -v 1 repack/oldpacks repack/freshened | LC_ALL=C sed 's/\.pack$//')"
1571 [ -z "$deadpacks" ] ||
1572 eval echo "$(printf '"%s".* ' $deadpacks)" | xargs rm -f || :
1574 # No need for this anymore
1575 rm -rf repack/alt objects/pack/repack
1576 unset GIT_ALTERNATE_OBJECT_DIRECTORIES
1579 ## Phase IV
1582 progress "~ [$proj] running gc prune-packed"
1584 # We do not want the redundant packs or any new "--keep-unreachable" pack(s) to be
1585 # present while running prune-packed. We try to guarantee that any loose object
1586 # (or any object present in a pack with an _l suffix which was created by mini gc)
1587 # that's unreachable persists for at least one $Girocco::Config::min_gc_interval
1588 # (not withstanding administrator interference to force earlier gc to occur).
1589 # If we were to include the redundant/keep-unreachable pack(s) when running
1590 # prune-packed and a loose unreachable object happened to be duplicated in one
1591 # of them we would end up removing it too soon and void our guarantee.
1592 git prune-packed $quiet
1594 progress "~ [$proj] running loose objects gc pack-objects ($(date))"
1596 # Although Git v2.10.0 and later support a --pack-loose-unreachable option,
1597 # we MUST NOT use it for these reasons:
1598 # 1) We're not interested in expensive "unreachable" at this point, only "loose"
1599 # 2) It produces simply horrid packs about 3.8x times larger than they should be
1600 # 3) We don't require anything more than Git v1.6.6
1601 # The only way we could see any _o pack files at this point is if one got
1602 # "freshened" while we were running gc. If that happens then it gets to live on
1603 # until the next full gc and we need to include it in the loose repack here.
1604 lpacks="$(list_packs --exclude-no-idx --exclude-no-sfx _l --exclude-no-sfx _o --quiet objects/pack |
1605 run_combine_packs --replace --names --loose --weak-naming --non-empty --honor-pack-keep \
1606 --all-progress-implied ${quiet:---progress} $packopts $newdeltas "$@")"
1608 if [ -n "$lpacks" ]; then
1609 # Make sure any primary pack(s) have a more recent mod time than "unreachable" objects packs
1610 if [ -n "$packs" ]; then
1611 sleep 1
1612 printf 'objects/pack/pack-%s_f.pack\n' $packs | xargs touch -c 2>/dev/null || :
1614 # We need to identify these packs later so we don't combine_packs them
1615 for objpack in $lpacks; do
1616 rename_pack "objects/pack/pack-$objpack" "objects/pack/pack-${objpack}_o" || :
1617 done
1620 # Polish up the final packs now
1621 rm -f objects/pack/pack-*_f.keep
1622 for pack in $packs $spacks; do
1623 rename_pack "objects/pack/pack-${pack}_f" "objects/pack/pack-$pack"
1624 done
1626 if [ -n "$lpacks" ]; then
1627 # Finally zap the corresponding loose objects
1628 progress "~ [$proj] running packed loose objects gc prune-packed"
1629 git prune-packed $quiet
1632 ! [ -e .gc_failed ] || exit 1
1633 # These, if they exist, are now meaningless and need to be removed
1634 rm -f gfi-packs .needsgc .needspack .needspackgc
1636 # Make sure this stays up to date
1637 git update-server-info
1639 # We must make loose objects group writable so that they
1640 # can be freshened by other pushers. Technically we need only do this for
1641 # push projects but to enable mirror projects to be more easily converted to
1642 # push projects, we go ahead and do it for all projects.
1643 # By the time we get here we really shouldn't have any of these, but just in case.
1644 { find -L objects/$octet -type f -name "$octet19*" -exec chmod ug+w '{}' + || :; } 2>/dev/null
1646 # darcs mirrors have a xxx.log file that will grow endlessly
1647 # if this is a mirror and the file exists, shorten it to 10000 lines
1648 # also take this opportunity to optimize the darcs repo
1649 if ! [ -e .nofetch ] && [ -n "$cfg_mirror" ]; then
1650 url="$(config_get baseurl)" || :
1651 case "$url" in darcs://* | darcs+http://* | darcs+https://*)
1652 if [ -n "$cfg_mirror_darcs" ]; then
1653 url="${url%/}"
1654 basedarcs="$(basename "${url#darcs*:/}")"
1655 if [ -f "$basedarcs.log" ]; then
1656 tail -n 10000 "$basedarcs.log" >"$basedarcs.log.$$"
1657 mv -f "$basedarcs.log.$$" "$basedarcs.log"
1659 if [ -d "$basedarcs.darcs" ]; then
1661 cd "$basedarcs.darcs"
1662 # without show_progress suppress non-error output
1663 [ "${show_progress:-0}" != "0" ] || exec >/dev/null
1664 # Note that this does not optimize _darcs/inventories/ :(
1665 darcs optimize || :
1669 esac
1672 # Create a matching .bndl header file for the all-in-one pack we just created
1673 # but only if we're not a fork (otherwise the bundle would not be complete)
1674 # and we are running at least Git version 1.7.2 (pack_is_complete always fails otherwise)
1675 if [ -n "$makebndl" ] && [ -n "$var_have_git_172" ]; then
1676 # There should only be one pack in $packs but do some checking...
1677 # The one we just created will have a .idx and will NOT have a .keep
1678 progress "~ [$proj] creating downloadable bundle header"
1679 pkbase=
1680 pkhead=
1681 IFS= read -r curhead <repack/HEAD.orig || :
1683 [ -s "objects/pack/pack-$packs.pack" ] &&
1684 [ -s "objects/pack/pack-$packs.idx" ] &&
1685 ! [ -e "objects/pack/pack-$packs.keep" ] &&
1686 pkhead="$(pack_is_complete "$PWD/objects/pack/pack-$packs.pack" \
1687 "$PWD/repack/packed-refs.orig" "$curhead")"
1688 then
1689 pkbase="objects/pack/pack-$packs"
1691 if [ -n "$pkbase" ] && [ -n "$pkhead" ]; then
1693 symref=
1694 case "$curhead" in "ref: refs/"?*|"ref:refs/"?*|"refs/"?*)
1695 symref="${curhead#ref:}"
1696 symref="${symref# }"
1697 esac
1698 bndlurl=
1699 [ -z "$cfg_httpbundleurl" ] || bndlurl=" url=$cfg_httpbundleurl/$proj.git/clone.bundle"
1700 echo "# v2 git bundle"
1701 LC_ALL=C sed -ne "/^$octet20$hexdig* refs\/[^ $tab]*\$/ p" <repack/packed-refs.orig
1702 if [ -n "$symref" ]; then
1703 printf "$pkhead HEAD\0symref=HEAD:%s%s\n" "$symref" "$bndlurl"
1704 else
1705 if [ -n "$bndlurl" ]; then
1706 printf "$pkhead HEAD\0%s\n" "${bndlurl# }"
1707 else
1708 echo "$pkhead HEAD"
1711 echo ""
1712 } >"$pkbase.bndl"
1713 bndletag="$("$cfg_basedir/bin/rangecgi" --etag -m 1 "$pkbase.bndl" "$pkbase.pack")" || :
1714 bndlsha="$(printf '%s' "$bndletag" | git hash-object --stdin)" || :
1715 if [ -n "$bndletag" ]; then
1716 case "$bndlsha" in $octet20*)
1717 bndlshatrailer="${bndlsha#????????}"
1718 bndlshaprefix="${bndlsha%$bndlshatrailer}"
1719 bndlname="$(TZ=UTC date +%Y%m%d_%H%M%S)-${bndlshaprefix:-0}"
1720 [ -d bundles ] || mkdir bundles
1721 echo "${pkbase#objects/pack/}.bndl" >"bundles/$bndlname"
1722 echo "${pkbase#objects/pack/}.pack" >>"bundles/$bndlname"
1723 ln -s -f -n "$bndlname" bundles/latest
1724 esac
1729 # Record the size of this repo as the sum of its clone packed-refs + *.pack sizes as 1024-byte blocks
1730 eval "reposizek=$(( $(
1731 echo 0 $(du -k repack/packed-refs.orig $(printf 'objects/pack/pack-%s.pack ' $packs) 2>/dev/null |
1732 LC_ALL=C awk '{print $1}') |
1733 LC_ALL=C sed -e 's/ / + /g') ))"
1734 config_set_raw girocco.reposizek "${reposizek:-0}"
1736 # Now we're finally done with this
1737 rm -rf repack
1739 # We didn't used to do anything about rerere or worktrees but we're
1740 # trying to make nice with linked working trees these days :)
1741 # Maybe even non-bare repositories too, but *shush* about those ;)
1742 if [ -n "$var_have_git_250" ] && [ -d worktrees ]; then
1743 # The value "3.months.ago" is hard-coded into gc.c rather than
1744 # having the default be in worktree.c so we must provide it if
1745 # we get nothing out of the gc.worktreePruneExpire config item
1746 # Prior to Git v2.6.0 the config item was gc.pruneworktreesexpire
1747 # however we just always use the newer name no matter what Git version
1748 expiry="$(git config --get gc.worktreePruneExpire 2>/dev/null)" || :
1749 eval git worktree prune --expire '"${expiry:-3.months.ago}"' "${quiet:+>/dev/null 2>&1}" || :
1751 # git rerere does it right and handles its own default/config'd expiration values
1752 ! [ -d rr-cache ] || eval git rerere gc "${quiet:+>/dev/null 2>&1}" || :
1754 # We use $gcstart here to avoid a race where a push occurs during the gc itself
1755 # and the next future gc could be incorrectly skipped if we used the current
1756 # timestamp here instead
1757 config_set lastgc "$gcstart"
1758 rm -f "$lockf"
1760 progress "- [$proj] garbage check ($(date))"