taskd/clone.sh: exponential backoff for git-svn retries
[girocco.git] / jobd / gc.sh
blobc3393f07f203e72fd203010d0691694572b2603d
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 # The pre-receive script creates one ref log file per push but we want them to
29 # be coalesced into one ref log file per day. We are guaranteed that any files
30 # we find to coalesce are NOT currently being written to since they are always
31 # written first as temporary files and then moved into place. We attempt to
32 # transfer the most recent modification time to the coalesced log file which
33 # would step on its mod time if it were being written to directly, but if we
34 # find per-process ref log files then it must be a push project and the only
35 # thing that would write directly to the main per-day log file would be a
36 # mirror project so there's actually no conflict.
37 # Also, if the clock is wonky (or was futzed with) we may have both YYYYMMDD
38 # and YYYYMMDD.gz present in which case combine them into YYYYMMDD
39 coalesce_reflogs() {
40 [ -d reflogs ] || return 0
41 rm -f .gc_failed
42 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 |
43 while read -r rname; do
44 if [ -e "$rname.gz" ]; then
45 if [ -s "$rname" ]; then
46 # Presumably the .gz file must have been created before the non-gz
47 # file since it had to be uncompressed at some point therefore
48 # we need to append the non-gz contents to it but keep the non-gz
49 # contents timestamp so we rename to YYYYMMDD_ which will sort first
50 # and be picked up in the next step if we are interrupted in the middle.
51 # If a YYYYMMDD_ file already exists we append to it and transfer the
52 # timestamp. Finally we transfer the YYYYMMDD_ timestamp to the result
53 # and remove the YYYYMMDD_ temporary file leaving the result uncompressed.
54 if [ -e "${rname}_" ]; then
55 cat "$rname" >>"${rname}_"
56 touch -r "$rname" "${rname}_"
57 rm -f "$rname"
58 ! [ -e "$rname" ]
59 else
60 mv "$rname" "${rname}_"
62 gzip -d "$rname.gz" </dev/null
63 [ -e "$rname" ] && ! [ -e "$rname.gz" ]
64 cat "${rname}_" >>"$rname"
65 touch -r "${rname}_" "$rname"
66 rm -f "${rname}_"
67 else
68 # Just remove the empty file to resolve the problem
69 rm -f "$rname"
72 done
73 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 |
74 while read -r rname; do
75 logname="${rname%%_*}"
76 # If someone's been futzing with the date, the file we want to
77 # append to could already have been compressed, so we just uncompress
78 # it here. The previous block guarantees we do not have both a compressed
79 # and uncompressed version present at the same time.
80 if [ -e "$logname.gz" ]; then
81 gzip -d "$logname.gz" </dev/null
82 [ -e "$logname" ] && ! [ -e "$logname.gz" ]
84 cat "$rname" >>"$logname"
85 touch -r "$rname" "$logname"
86 rm -f "$rname"
87 if [ -e "$rname" ]; then
88 >.gc_failed
89 echo "! [$proj] failed to remove $rname" >&2
90 exit 1 # will only exit subshell created by "|"
92 done
93 ! [ -e .gc_failed ]
96 # Remove any files in reflogs that are older than $cfg_reflogs_lifetime days
97 prune_reflogs() {
98 [ -d reflogs ] || return 0
99 exp="$(( ${cfg_reflogs_lifetime:-1} * 1440 ))"
100 [ $exp -gt 0 ] || exp=1440
101 [ $exp -le 43200 ] || exp=43200
102 find -L reflogs -maxdepth 1 -type f -mmin "+$exp" -exec rm -f '{}' + || :
105 # Compact any reflogs that are not today's UTC date unless a .gz version exists
106 compact_reflogs() {
107 [ -d reflogs ] || return 0
108 _td="reflogs/$(TZ=UTC date '+%Y%m%d')"
109 find -L reflogs -maxdepth 1 -type f -name "[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]" -print |
110 while read -r rname; do
111 [ "$rname" != "$_td" ] || continue
112 ! [ -e "$rname.gz" ] || continue
113 gzip -9 "$rname" </dev/null
114 done
117 # return true if there's more than one objects/pack-<sha>.pack file or
118 # ANY sha-1 files in objects or
119 # there's one pack and it's not a normal pack name or
120 # there's one pack but not any refs
121 is_dirty() {
122 _packs="$(find -L objects/pack -name "pre-auto-gc-[12].pack" -prune -o -name "*.pack" -type f -print 2>/dev/null | head -n 2)"
123 v_cnt _packscnt $_packs
124 if [ $_packscnt -gt 1 ]; then
125 return 0
127 if [ $_packscnt -eq 1 ]; then
128 # the single pack name is in $_packs
129 _packs="${_packs%.pack}"
130 _packs="${_packs#objects/pack/}"
131 case "$_packs" in
132 pack-*)
133 _packs="${_packs#pack-}"
134 if [ "${#_packs}" -lt 40 ] || [ "${_packs#*[!0-9a-fA-F]}" != "$_packs" ]; then
135 # name not exclusively 40 or more hexadecimal digits makes it dirty
136 return 0
140 # abnormal name makes it dirty
141 return 0
143 esac
145 _objs=$(find -L objects/$octet -name "$octet19*" -type f -print 2>/dev/null | head -n 1 | LC_ALL=C wc -l)
146 [ $_objs -eq 0 ] || return 0
147 [ $_packscnt -eq 1 ] || return 1
148 # we do this check last because it's potentially the most expensive;
149 # at this point we know we do not have any loose objects, but we do
150 # have one pack that's named "normally"; empty refs => dirty
151 is_empty_refs_dir
154 # combine the input pack(s) into a new pack (or possibly packs if packSizeLimit set)
155 # input pack names are read from standard input one per line delimited by the first
156 # ':', ' ' or '\n' character on the line (which allows gfi-packs to be read directly)
157 # all arguments, if any, are passed to pack-objects as additional options
158 # returns non-zero on failure AND creates .gc_failed in that case
159 combine_packs() {
160 rm -f .gc_failed
161 combine_packs_std "$@" || {
162 >.gc_failed
163 return 1
165 return 0
168 # if the current directory is_gfi_mirror then repack all packs listed in gfi-packs
169 repack_gfi_packs() {
170 [ -n "$gfi_mirror" ] || return 0
171 [ -d objects/pack ] || { rm -f gfi-packs; return 0; }
172 progress "~ [$proj] redeltifying poor quality git fast-import packs"
173 combine_packs --ignore-missing --no-reuse-delta <gfi-packs
174 rm -f gfi-packs
175 return 0
178 # pack any existing loose objects into a new _l.pack file then run prune-packed
179 # note that prune-packed is NOT run beforehand -- the caller must do that if needed
180 # loose objects need not be part of complete commits/trees as --weak-naming is used
181 pack_loose_objects() {
182 _lpacks="$(run_combine_packs </dev/null --names --loose --weak-naming --non-empty --all-progress-implied ${quiet:---progress} $packopts)"
183 if [ -n "$_lpacks" ]; then
184 # We need to identify these packs later so we don't combine_packs them
185 for _objpack in $_lpacks; do
186 rename_pack "objects/pack/pack-$_objpack" "objects/pack/pack-${_objpack}_l" || :
187 done
188 git prune-packed $quiet
192 # combine small packs into larger pack(s)
193 # we avoid any _[lo], keep, bndl or bitmap packs
194 # if the optional argument is non-empty even a single small pack will be redeltad
195 combine_small_packs() {
196 _didprogress=
197 _minsmallpacks=2
198 if [ -n "$1" ] && [ -n "$noreusedeltaopt" ]; then
199 _minsmallpacks=1
201 _lpo="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl"
202 _lpo="$_lpo --exclude-sfx _u --exclude-sfx _o --exclude-sfx _l"
203 _lpo="$_lpo --quiet --object-limit $var_redelta_threshold objects/pack"
204 while
205 _cnt="$(list_packs --count $_lpo)" || :
206 test "${_cnt:-0}" -ge $_minsmallpacks
208 [ -n "$_didprogress" ] || {
209 progress "~ [$proj] combining small packs into a single larger pack"
210 _didprogress=1
212 _newp="$(list_packs $_lpo | combine_packs --names $noreusedeltaopt)"
213 v_cnt _newc $_newp
214 # be paranoid and exit the loop if we haven't reduced the number of packs
215 [ $_newc -lt $_cnt ] || break
216 _minsmallpacks=2
217 done
218 return 0
221 # combine small _l packs into larger pack(s) using --weak-naming
222 # we avoid any non _l, keep, bndl or bitmap packs
223 # if the optional 2nd argument is non-empty even a single small pack will be redeltad
224 combine_small_loose_packs() {
225 _didprogress=
226 _minsmallpacks=2
227 if [ -n "$1" ] && [ -n "$noreusedeltaopt" ]; then
228 _minsmallpacks=1
230 _lpo="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl"
231 _lpo="$_lpo --exclude-no-sfx _l"
232 _lpo="$_lpo --quiet --object-limit $var_redelta_threshold objects/pack"
233 while
234 _cnt="$(list_packs --count $_lpo)" || :
235 test "${_cnt:-0}" -ge $_minsmallpacks
237 [ -n "$_didprogress" ] || {
238 progress "~ [$proj] combining small loose packs into a single larger pack"
239 _didprogress=1
241 _newp="$(list_packs $_lpo | combine_packs --names --weak-naming $noreusedeltaopt)"
242 # We need to identify these packs later so we don't combine_packs them
243 for _objpack in $_newp; do
244 rename_pack "objects/pack/pack-$_objpack" "objects/pack/pack-${_objpack}_l" || :
245 done
246 v_cnt _newc $_newp
247 # be paranoid and exit the loop if we haven't reduced the number of packs
248 [ $_newc -lt $_cnt ] || break
249 _minsmallpacks=2
250 done
251 return 0
254 # Unfortunately, some fetch strategies (e.g. git-svn and non-smart HTTP) lack
255 # the ability to store newly fetched objects in a pack.
256 # However, the fetch code conveniently sets .needspack just before it fetches
257 # so that it's easy to find all the loose objects that have been fetched and
258 # combine them into a pack. The --no-reuse-delta option is meaningless here
259 # since everything to be packed is a loose object and therefore not a delta so
260 # deltification will always take place.
261 make_needs_pack() {
262 [ -f .needspack ] || return 0
263 rm -f .needspackgc
264 mv -f .needspack .needspackgc
265 progress "~ [$proj] combining fetched loose objects into a pack"
266 _newp="$(find -L objects/$octet -maxdepth 1 -type f -newer .needspackgc -name "$octet19*" -print 2>/dev/null |
267 LC_ALL=C awk -F / '{print $2 $3}' |
268 run_combine_packs --objects --names $packopts --incremental --all-progress-implied $quiet --non-empty)" || {
269 # We used to fail gc here.
270 # Now, however, we just ignore the failure because we have
271 # another mechanism to handle loose objects and it's possible
272 # that the fetcher somehow brought in unconnected objects which
273 # would cause the above combine-packs to fail.
274 # By ignoring the failure and just removing the .needspack file
275 # the loose objects will be treated as "ordinary" loose objects
276 # and packed using the "--weak-naming" option which can handle
277 # broken connectivity.
278 # That's a better solution than just failing here or leaving
279 # .needspack behind to potentially continue to fail again and
280 # again.
281 _newp=
283 if [ -n "$_newp" ]; then
284 # remove the now-redundant loose objects -- this is always safe
285 # even during a concurrent push because a reprepare_packed_git
286 # will be triggered if an object that should be there is not
287 # found thereby finding it in the new pack instead
288 git prune-packed $quiet
290 rm -f .needspackgc
293 # HEADSHA="$(pack_is_complete /full/path/to/some.pack /full/path/to/packed-refs "$(cat HEAD)")"
294 pack_is_complete() {
295 # Must have a matching .idx file and a non-empty packed-refs file
296 [ -s "${1%.pack}.idx" ] || return 1
297 [ -s "$2" ] || return 1
298 _headsha=
299 case "$3" in
300 $octet20*)
301 _headsha="$3"
303 "ref: refs/"?*|"ref:refs/"?*|"refs/"?*)
304 _headmatch="${3#ref:}"
305 _headmatch="${_headmatch# }"
306 _headmatchpat="$(echo "$_headmatch" | LC_ALL=C sed -e 's/\([.$]\)/\\\1/g')"
307 _headsha="$(LC_ALL=C grep -e "^$octet20$hexdig* $_headmatchpat\$" <"$2" |
308 LC_ALL=C cut -d ' ' -f 1)"
309 case "$_headsha" in $octet20*) :;; *)
310 return 1
311 esac
314 # bad HEAD
315 return 1
316 esac
317 rm -rf pack_is_complete_test
318 mkdir pack_is_complete_test
319 mkdir pack_is_complete_test/refs
320 mkdir pack_is_complete_test/objects
321 mkdir pack_is_complete_test/objects/pack
322 echo "$_headsha" >pack_is_complete_test/HEAD
323 ln -s "$1" pack_is_complete_test/objects/pack/
324 ln -s "${1%.pack}.idx" pack_is_complete_test/objects/pack/
325 ln -s "$2" pack_is_complete_test/packed-refs
326 _count="$(git --git-dir=pack_is_complete_test rev-list --count --all 2>/dev/null)" || :
327 rm -rf pack_is_complete_test
328 [ -n "$_count" ] || return 1
329 [ "$_count" -gt 0 ] 2>/dev/null || return 1
330 echo "$_headsha"
333 # On return a "$lockf" will have been created that must be removed when gc is done
334 lock_gc() {
335 v_lock_gc _lockresult || {
336 echo >&2 "[$proj] $_lockresult"
337 exit 1
339 lockf="$_lockresult"
342 # Create a repack subdirectory such that running repack in it will pack the
343 # same things that a pack in the normal directory would except that the pack
344 # is guaranteed to be generated in an optimized order by adding a suitable
345 # synthesized ref in the refs/tags namespace (yes, pack-objects.c really does
346 # behave differently depending on the contents of the refs/tags namespace).
347 # Before calling this, pack-refs --all MUST be performed or the wrong pack
348 # will end up being made.
350 # If a ref deletion is pushed after making the repack subdir but before the
351 # the actual repack, the discarded objects will be packed -- no big deal,
352 # they'll get discarded the next time gc runs.
354 # If a fast-forward ref update is pushed after making the repack subdir but
355 # before the actual repack, it will be picked up and the new objects packed
356 # (subject to the normal git repack race about picking such updates up).
358 # If a non-fast-forward ref update is pushed after making the repack subdir but
359 # before the actual repack, it will be picked up like a fast-forward update but
360 # the discarded objects will be included like a ref deletion (until the next
361 # scheduled gc takes place).
363 # We retain a copy of the original packed-refs file as repack/packed-refs.orig
364 # If ref deletions come in while we're repacking, the original packed-refs
365 # file will be modified, but we'll still pack the deleted ref(s).
366 # If the packed-refs.orig file is used to create the bundle header we avoid
367 # a situation where the bundle contains a ref state that never actually
368 # existed in reality (for example a new branch is pushed and then an old
369 # branch deleted afterwards -- the deletion would show up in the bundle
370 # because it will cause the original packed-refs file to be re-written, but
371 # the new branch creation will not unless we do another pack-refs which might
372 # lead to having in incomplete bundle). Therefore we want to keep a copy of
373 # the original packed-refs file around. We do the same thing for HEAD.
375 # It's possible that the "objects" subdirectory is a symbolic link.
376 # Git does support this. However, during the repacking process, new packs
377 # will be created in repack/alt/pack and then moved into objects/pack.
378 # In order for this to work seemlessly, they must both be on the same
379 # filesystem. But when objects (or even objects/pack) is a symbolic link they
380 # might not be. For this reason a "repack" subdirectory is created under
381 # objects/pack and the repack/alt/pack directory symbolicly linked to it.
383 # Git allows not just HEAD to be a symbolic-ref, but any ref anywhere in the
384 # refs namespace. We are concerned about ref name collisions and getting the
385 # right tag set to get an optimal pack. We can safely duplicate the ref space
386 # under refs/heads, refs/notes and refs/remotes without any risk of unwanted
387 # collisions and this will likely make over 99%+ of all symbolic refs found
388 # in the wild work properly. Girocco itself never creates any symbolic refs
389 # inside the refs namespace; this is a nod to simultaneously using a Girocco
390 # repository for other purposes.
391 make_repack_dir() {
392 ! [ -d repack ] || rm -rf repack
393 ! [ -d repack ] || { echo >&2 "[$proj] cannot remove repack subdirectory"; exit 1; }
394 [ -d objects/pack ] || mkdir -p objects/pack
395 ! [ -d objects/pack/repack ] || rm -rf objects/pack/repack
396 ! [ -d objects/pack/repack ] || { echo >&2 "[$proj] cannot remove objects/pack/repack subdirectory"; exit 1; }
397 mkdir repack repack/refs repack/alt objects/pack/repack
398 [ -d info ] || mkdir info
399 ln -s ../config repack/config
400 ln -s ../info repack/info
401 ln -s ../objects repack/objects
402 ln -s "$PWD/objects/pack/repack" repack/alt/pack
403 ln -s ../../refs repack/refs/refs
404 ! [ -d logs ] || ln -s ../logs repack/logs
405 ! [ -d worktrees ] || ln -s ../worktrees repack/worktrees
406 _lines=$(( $(LC_ALL=C wc -l <packed-refs) ))
407 cat HEAD >repack/HEAD.orig
408 >repack/packed-refs.extra
409 _xtralines=0
410 cat packed-refs >repack/packed-refs.orig
411 if [ $(LC_ALL=C wc -l <repack/packed-refs.orig) -ne "$_lines" ]; then
412 echo >&2 "[$proj] error: make_repack_dir failed original packed-refs line count sanity check"
413 exit 1
415 if [ "${cfg_fetch_stash_refs:-0}" = "0" ]; then
416 # migrate any refs/stash or refs/tgstash lines to repack/packed-refs.extra
417 <repack/packed-refs.orig LC_ALL=C awk -v xtra="repack/packed-refs.extra" '
418 BEGIN { peeling = 0 }
419 NR == 1 && /^#/ { print; next; }
420 peeling && /^\^/ { print >>xtra; next; }
421 /^[0-9A-Fa-f][0-9A-Fa-f][0-9A-Fa-f][0-9A-Fa-f]+ refs\/(stash|tgstash)(\/|$)/ {
422 peeling = 1
423 print >>xtra
424 next
426 { peeling = 0; print; }
427 ' >repack/packed-refs.new
428 _xtralines="$(( $(LC_ALL=C wc -l <repack/packed-refs.extra) + 0 ))"
429 _newlines="$(( $(LC_ALL=C wc -l <repack/packed-refs.new) + 0 ))"
430 if [ "$(( $_newlines + $_xtralines ))" -ne "$_lines" ]; then
431 echo >&2 "[$proj] error: make_repack_dir failed packed-refs.extra line count sanity check"
432 exit 1
434 mv -f repack/packed-refs.new repack/packed-refs.orig
435 _lines="$_newlines"
437 # Note: Git v1.5.0 introduced the "# pack-refs with:" header line for the packed-refs file
438 sed '/^# pack-refs/d; s, refs/, refs/!/,' <repack/packed-refs.orig >repack/packed-refs
439 nohead=
440 headref="$(git rev-parse --verify --quiet HEAD)" || :
441 if [ -n "$headref" ]; then
442 echo "$headref refs/!=/HEAD" >>repack/packed-refs
443 echo "$headref refs/heads/!" >>repack/packed-refs
444 nohead='\, refs/heads/!$,d; '
445 _lines=$(( $_lines + 2 ))
447 if [ $(( $(LC_ALL=C wc -l <repack/packed-refs) + 1 )) -ne "$_lines" ]; then
448 echo >&2 "[$proj] error: make_repack_dir failed packed-refs initial line count sanity check"
449 exit 1
451 sed -n "$nohead"'\, refs/heads/,p; \, refs/notes/,p; \, refs/remotes/,p' <repack/packed-refs.orig >>repack/packed-refs
452 _newlines="$(( $(LC_ALL=C wc -l <repack/packed-refs) ))"
453 if [ $(( $_newlines + 1 )) -lt "$_lines" ]; then
454 echo >&2 "[$proj] error: make_repack_dir failed packed-refs extra line count sanity check"
455 exit 1
457 _lines="$_newlines"
458 optref="$(git rev-list -n 1 --all 2>/dev/null)" || :
459 if [ -n "$optref" ]; then
460 echo "$optref refs/tags/!" >>repack/packed-refs
461 _lines=$(( $_lines + 1 ))
462 echo "$optref" >repack/HEAD
463 else
464 cat HEAD >repack/HEAD
466 if [ $(LC_ALL=C wc -l <repack/packed-refs) -ne "$_lines" ]; then
467 echo >&2 "[$proj] error: make_repack_dir failed packed-refs line count sanity check"
468 exit 1
472 # Remove any crud that's been left behind by interrupted operations
473 # that did not clean up after themselves
474 remove_crud() {
475 # Remove any existing FETCH_HEAD
476 # There can only be a FETCH_HEAD if we've been fetching, not if we've been
477 # receiving pushes (those never create a FETCH_HEAD).
478 # And if we're fetching because we're a mirror, we know we're not fetching right
479 # now since jobd.pl never runs a project's fetch simultaneously with its gc.
480 # Therefore any existing FETCH_HEAD is junk. And it may be many megabytes if
481 # there were a lot of refs.
482 rm -f FETCH_HEAD
484 # remove any existing pack_is_complete_test or repack subdirectories
485 # If either exists when this function is called it's crud
486 rm -rf pack_is_complete_test repack objects/pack/repack
488 # Remove any stale pack remnants that are more than an hour old.
489 # Stale pack fragments are defined as any pack-<sha1>.ext where .ext is NOT
490 # .pack AND the corresponding .pack DOES NOT exist. A bunch of stale
491 # pack-<sha1>.idx files without their corresponding .pack files are worthless
492 # and just waste space. Normally there shouldn't be any remnants but actually
493 # this can happen when things are interrupted at just the wrong time.
494 # Note that the objects/pack directory is created by git init and should
495 # always exist.
496 find -L objects/pack -maxdepth 1 -type f -mmin +60 -name "pack-$octet20*.?*" -print |
497 LC_ALL=C sed -e 's/^objects\/pack\/pack-//; s/\..*$//' | LC_ALL=C sort -u |
498 while read packsha; do
499 ! [ -e "objects/pack/pack-$packsha.pack" ] || continue
500 rm -f "objects/pack/pack-$packsha".?*
501 done
503 # Remove any stale tmp reflogs files that are more than one hour old.
504 # Since they are created only while the pre-receive hook is running and
505 # all it does is process a bunch of refs passed to it on standard input
506 # it's inconceivable that it would ever take as much as an hour to run.
507 if [ -d reflogs ]; then
508 find -L reflogs -maxdepth 1 -type f -mmin +60 -name "tmp_*" -exec rm -f '{}' + || :
511 # Remove any stale object tmp_obj_* files that are more than 3 hours old.
512 # Really these files should only exist very briefly so there shouldn't be any
513 # but things happen that can end up leaving them behind.
514 find -L objects/$octet -maxdepth 1 -type f -mmin +180 -name "tmp_obj_?*" -exec rm -f '{}' + 2>/dev/null || :
516 # Remove any stale pack .keep files that are more than 12 hours old.
517 # We don't do anything to create any permanent pack .keep files, so they must
518 # be remnants from some failed push or something. Removing the .keep will
519 # allow the pack to be properly repacked.
520 find -L objects/pack -maxdepth 1 -type f -mmin +720 -name "pack-$octet20*.keep" -exec rm -f '{}' + || :
522 # Remove any stale tmp_pack_*, tmp_idx_*, tmp_bitmap_*, packtmp-* or .tmp-*-pack* files
523 # that are more than 12 hours old.
524 find -L objects/pack -maxdepth 1 -type f -mmin +720 \( \
525 -name "tmp_pack_?*" -o -name "tmp_idx_?*" -o -name "tmp_bitmap_?*" -o \
526 -name "packtmp-?*" -o -name ".tmp-?*-pack*" \
527 \) -exec rm -f '{}' + || :
529 # Remove any stale incoming-* object quarantine directories that are
530 # more than 12 hours old. These are new with Git >= 2.11.0.
531 find -L objects -maxdepth 1 -type d -name 'incoming-?*' -mmin +720 \
532 -exec rm -rf '{}' + || :
534 # Remove any stale shallow_* files that are more than 12 hours old.
535 # These can be left behind by Git >= 1.8.4.2 and < 2.0.0 when a client
536 # requests a shallow clone. Also discard stale .refs-temp* and
537 # .refs-new* files at the same time.
538 find -L . -maxdepth 1 -type f -mmin +720 \( \
539 -name "shallow_?*" -o -name ".refs-temp*" -o -name ".refs-new*" \
540 \) -exec rm -f '{}' + || :
542 # Remove any stale cmbnpcks-* dirs that are more than 12 hours old.
543 # These can be left behind by abnormal exits (e.g. power failure).
544 find -L . -maxdepth 1 -type d -mmin +720 -name "cmbnpcks-?*" \
545 -exec rm -rf '{}' + || :
547 # Remove any stale *.temp files in the objects area that are more than 12 hours old.
548 # This can be stale sha1.temp, or stale *.pack.temp so we kill all stale *.temp.
549 find -L objects -type f -mmin +720 -name "*.temp" -exec rm -f '{}' + || :
551 # Remove any stale *.lock files in the htmlcache area that might have been left
552 # behind after an abnormal exit during an attempt to update a cached file and
553 # are more than 1 hour old.
554 ! [ -d htmlcache ] || find -L htmlcache -type f -mmin +60 -name "*.lock" -exec rm -f '{}' + || :
556 # Remove any stale git-svn temp files that are more than 12 hours old.
557 # The git-svn process creates temp files with random 10 character names
558 # in the root of $GIT_DIR. Unfortunately they do not have a recognizable
559 # prefix, so we just have to kill any files with a 10-character name. We
560 # do this only for git-svn mirrors. All characters are chosen from
561 # [A-Za-z0-9_] so we can at least check that and fortunately the only
562 # collision is 'FETCH_HEAD' but that shouldn't matter.
563 # There may also be temp files with a Git_ prefix as well.
564 if [ -n "$svn_mirror" ]; then
565 _randchar='[A-Za-z0-9_]'
566 _randchar2="$_randchar$_randchar"
567 _randchar4="$_randchar2$_randchar2"
568 _randchar10="$_randchar4$_randchar4$_randchar2"
569 find -L . -maxdepth 1 -type f -mmin +720 -name "$_randchar10" -exec rm -f '{}' + || :
570 find -L . -maxdepth 1 -type f -mmin +720 -name "Git_*" -exec rm -f '{}' + || :
573 # Remove any stale fast_import_crash_<pid> files that are more than 3 days old.
574 if [ -n "$gfi_mirror" ]; then
575 find -L . -maxdepth 1 -type f -mmin +4320 -name "fast_import_crash_?*" -exec rm -f '{}' + || :
578 # Remove any stale core or *.core or core.* files that are more than 3 days old.
579 find -L . -maxdepth 1 -type f -mmin +4320 \( -name "core" -o -name "*.core" -o -name "core.*" \) \
580 -exec rm -f '{}' + || :
584 ## Garbage Collection Types
586 ## There are two kinds of possible garbage collection (gc) operations:
588 ## 1. A normal, full gc
589 ## 2. A "mini" gc
591 ## If the full garbage collection interval has expired (or gc has never been
592 ## run), then a normal, full gc will take place. Otherwise, a "mini" gc will
593 ## take place if the file .needsgc exists.
595 ## A "mini" gc is similar to "git gc --auto" in that it may not end up actually
596 ## doing anything unless the right conditions are present so it's not a burden
597 ## to run it often. If the file .needsgc exists, a "mini" gc will occur at
598 ## the next opportunity.
600 ## See the docs/technical/gc.txt and docs/technical/gc-mini.txt files for more
601 ## of the gory details of how garbage collection is performed.
603 ## Note, however, that the .nogc file suppresses ALL gc activity (normal or mini).
606 proj="${1%.git}"
607 shift
608 cd "$cfg_reporoot/$proj.git"
609 [ -d objects/pack ] || { rm -f gfi-packs; mkdir -p objects/pack; }
610 mirror_url="$(get_mirror_url)" || :
611 svn_mirror=
612 ! is_svn_mirror_url "$mirror_url" || svn_mirror=1
613 gfi_mirror=
614 if [ -f gfi-packs ] && [ -s gfi-packs ] && is_gfi_mirror_url "$mirror_url"; then
615 gfi_mirror=1
618 # If git config --bool --get girocco.redelta is explicitly false then automatic
619 # redelta when there are less than $var_redelta_threshold objects will be suppressed.
620 # On the other hand, if git config --get girocco.redelta is "always" then, on a full
621 # gc only, for the final repack, deltas will always be recomputed.
622 # This can be set on a per-project basis to avoid unusual pathological gc behavior.
623 # Setting this will hurt efficiency of the affected repository.
624 # Note that fast-import packs ALWAYS get new deltas regardless of this setting.
625 noreusedeltaopt="--no-reuse-delta"
626 [ "$(git config --bool --get girocco.redelta 2>/dev/null || :)" != "false" ] || noreusedeltaopt=
627 alwaysredelta=
628 [ "$(git config --get girocco.redelta 2>/dev/null || :)" != "always" ] || alwaysredelta=1
630 # Extract any -f or -F or --no-reuse-object or --no-reuse-delta options
631 # to be compatible with the old and new gc.sh versions and avoid ugly argument
632 # duplication in process lists at the same time
633 # Any options found will override the "girocco.redelta" setting
634 recompress=
635 idx=$#
636 while [ $idx -gt 0 ]; do
637 idx=$(( $idx - 1 ))
638 opt="$1"
639 shift
640 case "$opt" in
641 -f|--no-reuse-delta)
642 alwaysredelta=1
643 continue
645 -F|--no-reuse-object)
646 alwaysredelta=1
647 recompress=1
648 continue
650 -?*)
653 printf >&2 '%s\n' "bad non-option argument: $opt"
654 echo >&2 "(Did you perhaps intend to use a --xxx=yyy form?)"
655 exit 1
656 esac
657 [ -z "$opt" ] || set -- "$@" "$opt"
658 done
659 if [ -n "$alwaysredelta" ]; then
660 noreusedeltaopt="--no-reuse-delta"
661 [ -z "$recompress" ] || noreusedeltaopt="--no-reuse-object"
664 trap 'e=$?; rm -f .gc_in_progress; if [ $e != 0 ]; then echo "gc failed dir: $PWD" >&2; fi' EXIT
665 trap 'exit 130' INT
666 trap 'exit 143' TERM
668 # date -R is linux-only, POSIX equivalent is '+%a, %d %b %Y %T %z'
669 datefmt='+%a, %d %b %Y %T %z'
671 isminigc=
672 if [ "${force_gc:-0}" = "0" ] && check_interval lastgc $cfg_min_gc_interval; then
673 if [ -e .needsgc ]; then
674 isminigc=1
675 else
676 progress "= [$proj] garbage check skip (last at $(config_get lastgc))"
677 exit 0
680 if [ -e .nogc ]; then
681 progress "x [$proj] garbage check disabled"
682 exit 0
684 if ! [ -e .nofetch ] && [ -e .clone_in_progress ] && ! [ -e .clone_failed ]; then
685 progress "x [$proj] garbage check disabled (clone in progress)"
686 exit 0
688 if [ -z "$isminigc" ] && [ -e .delaygc ] && [ -e .needsgc ]; then
689 # Eligible for a full gc but .delaygc is set so it would be skipped
690 # However .needsgc is also set so transform it into a mini instead
691 isminigc=1
692 progress "~ [$proj] garbage check delayed but checking mini because .needsgc"
695 if [ -n "$isminigc" ]; then
696 # Perform a "mini" gc
697 # Note that .delaygc is ignored here as that's only intended for full gc
698 lock_gc
699 rm -f .allowgc .needsgc
700 rm -f objects/pack/pack-*_[rful].keep
701 remove_crud
702 coalesce_reflogs
703 prune_reflogs
704 compact_reflogs
705 maintain_auto_gc_hack
706 generate_auto_gc_update
707 miniactive=
708 if [ -f .needspack ]; then
709 miniactive=1
710 progress "+ [$proj] mini garbage check ($(date))"
711 make_needs_pack
713 if [ -z "$cfg_delay_gfi_redelta" ] && [ -n "$gfi_mirror" ]; then
714 # $Girocco::Config::delay_gfi_redelta is false, force redeltification now
715 if [ -z "$miniactive" ]; then
716 miniactive=1
717 progress "+ [$proj] mini garbage check ($(date))"
719 repack_gfi_packs
721 if lotsa_loose_objects; then
722 if [ -z "$miniactive" ]; then
723 miniactive=1
724 progress "+ [$proj] mini garbage check ($(date))"
726 pack_loose_objects
728 # If there aren't at least 10 non-keep, non-bitmap, non-bndl packs then
729 # don't actually process them yet
730 lpo="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl --quiet"
731 packcnt="$(list_packs --count $lpo objects/pack)" || :
732 if [ "${packcnt:-0}" -ge 10 ]; then
733 if [ -z "$miniactive" ]; then
734 miniactive=1
735 progress "+ [$proj] mini garbage check ($(date))"
737 # if we have at least 10 packs go ahead and pack all refs now too
738 git pack-refs --all --prune
739 if [ -n "$gfi_mirror" ]; then
740 repack_gfi_packs
741 packcnt="$(list_packs --count $lpo objects/pack)" || :
743 # if repack_gfi_packs dropped the pack count to < 10 don't combine
744 if [ "${packcnt:-0}" -ge 10 ]; then
745 combine_small_packs
746 combine_small_loose_packs
747 packcnt="$(list_packs --count $lpo objects/pack)" || :
749 # if we still have more than 10 packs trigger a full gc
750 if [ "${packcnt:-0}" -ge 10 ]; then
751 # We shouldn't be in a .delaygc state at this point, but if
752 # we are then nuke it because we really need a full gc now
753 rm -f .delaygc
754 git config --unset gitweb.lastgc
755 rm -f "$lockf"
756 git update-server-info # just in case
757 progress "- [$proj] mini garbage check triggering full gc too many packs ($(date))"
758 exit 0
761 rm -f "$lockf"
762 if [ -n "$miniactive" ]; then
763 git update-server-info
764 progress "- [$proj] mini garbage check ($(date))"
765 else
766 progress "= [$proj] mini garbage check nothing but crud removal to do ($(date))"
768 exit 0
771 # Avoid unnecessary garbage collections:
772 # 1. If lastreceive is set and is older than lastgc
773 # -AND-
774 # 2. We are not a fork (is_empty_alternates_file) -OR- lastparentgc is older than lastgc
776 # If lastgc is NOT set or lastreceive is NOT set we MUST run gc
777 # If we are a fork and lastparentgc is NOT set we MUST run gc
779 # If the repo is dirty after removing any crud we MUST run gc
781 gcstart="$(date "$datefmt")"
782 skipgc=
783 isfork=
784 is_empty_alternates_file objects/info/alternates || isfork=1
785 lastparentgcsecs=
786 [ -z "$isfork" ] || lastparentgcsecs="$(config_get_date_seconds lastparentgc)" || :
787 lastreceivesecs=
788 if lastreceivesecs="$(config_get_date_seconds lastreceive)" &&
789 [ "${force_gc:-0}" = "0" ] &&
790 lastgcsecs="$(config_get_date_seconds lastgc)" &&
791 [ $lastreceivesecs -lt $lastgcsecs ]; then
792 # We've run gc since we last received, so maybe we can skip,
793 # check if not fork or fork and lastparentgc < lastgc
794 if [ -n "$isfork" ]; then
795 if [ -n "$lastparentgcsecs" ] &&
796 [ $lastparentgcsecs -lt $lastgcsecs ]; then
797 # We've run gc since our parent ran gc so we can skip
798 skipgc=1
800 else
801 # We don't have any alternates (we're not a forK) so we can skip
802 skipgc=1
806 # Prevent any other simultaneous gc operations
807 lock_gc
809 # At this point, if .allowgc or .gc_failed exists, it's now crud to be removed
810 rm -f .allowgc .gc_failed
812 # Ideally we would do this in post-receive, but that would mean duplicating the
813 # logic so it's available in the chroot jail and that's highly undesirable
814 # Instead, since the first gc will be triggered immediately following the first
815 # push, we do the check here as it's quick and harmless if HEAD is already valid
816 check_and_set_head || :
818 # Always get rid of crud
819 remove_crud
821 # Always perform reflogs maintenance
822 coalesce_reflogs
823 prune_reflogs
824 compact_reflogs
826 # Always maintain auto gc hack
827 maintain_auto_gc_hack
828 generate_auto_gc_update
830 # Run 'git svn gc' now for svn mirrors
831 if [ -n "$svn_mirror" ]; then
832 git svn gc || :
835 # Skip the actual gc if .delaygc is set
836 if [ -e .delaygc ]; then
837 progress "x [$proj] garbage check delayed (except for crud removal)"
838 rm -f "$lockf"
839 exit 0
842 # Do not skip gc if the repo is dirty
843 if [ -n "$skipgc" ] && ! is_dirty; then
844 progress "= [$proj] garbage check nothing but crud removal to do ($(date))"
845 config_set lastgc "$gcstart"
846 rm -f "$lockf"
847 exit 0
850 bumptime=
851 if [ -n "$isfork" ] && [ -z "$lastparentgcsecs" ]; then
852 # set lastparentgc and then update gcstart to be at least 1 second later
853 config_set lastparentgc "$gcstart"
854 bumptime=1
856 if [ -z "$lastreceivesecs" ]; then
857 # set lastreceive and then update gcstart to be at least 1 second later
858 config_set lastreceive "$gcstart"
859 bumptime=1
861 if [ -n "$bumptime" ]; then
862 sleep 1
863 gcstart="$(date "$datefmt")"
866 progress "+ [$proj] garbage check ($(date))"
868 newdeltas=
869 [ -z "$alwaysredelta" ] || newdeltas="$noreusedeltaopt"
870 if [ -z "$newdeltas" ] && [ -n "$gfi_mirror" ]; then
871 if [ $(list_packs --exclude-no-idx --count objects/pack) -le \
872 $(list_packs --exclude-no-idx --count --quiet --only gfi-packs) ]; then
873 # Don't bother with repack_gfi_packs since everything's being repacked
874 newdeltas="--no-reuse-delta"
877 if [ -z "$newdeltas" ] && [ -n "$noreusedeltaopt" ] &&
878 [ $(list_packs --all --exclude-no-idx --count-objects objects/pack) -le $var_redelta_threshold ]; then
879 # There aren't enough objects to worry about so just redelta to get the best pack
880 newdeltas="--no-reuse-delta"
882 if [ -z "$newdeltas" ]; then
883 # Since we're not going to recompute deltas overall, we need to do the
884 # "mini" maintenance so that we can get more optimal deltas
885 [ -z "$noreusedeltaopt" ] || make_needs_pack
886 repack_gfi_packs
887 force_single_pack_redelta=
888 [ -n "$gfi_mirror" ] || [ -n "$svn_mirror" ] || force_single_pack_redelta=1
889 [ -z "$noreusedeltaopt" ] || combine_small_packs $force_single_pack_redelta
890 [ -z "$noreusedeltaopt" ] || combine_small_loose_packs $force_single_pack_redelta
894 ## Safe Pruning In Forks
896 ## We are about to perform garbage collection. We do NOT use the "git gc" or
897 ## the "git repack" commands directly as they do not provide enough control over
898 ## the fine details. However, we DO maintain a "gc.pid" file during our garbage
899 ## collection so that a simultaneous "git gc" by an administrator will be
900 ## blocked (and similarly we refuse to start garbage collection if we cannot
901 ## create the "gc.pid" file).
903 ## When we say "gc" in the below description we are referring to our "gc.sh"
904 ## script, NOT the "git gc" command.
906 ## If the project we are running garbage collection (gc) on has any forks we
907 ## must be careful not to remove any objects that while no longer referenced by
908 ## this project (the parent) are still referenced by one or more forks (the
909 ## children) otherwise the children will become corrupt and we can't abide
910 ## corrupt children.
912 ## One way to accomplish this is to simply hard-link all currently existing
913 ## loose objects and packs in the parent into all the children that refer to the
914 ## parent (via a line in their objects/info/alternates file) before beginning
915 ## the gc operation and then relying on a subsequent gc in the child to clean up
916 ## any excess objects/packs. We used to use this strategy but it's very
917 ## inefficient because:
919 ## 1. The disk space used by the old pack(s)/object(s) will not be reclaimed
920 ## until all children (and their children, if any) run gc by which time
921 ## it's quite possible the topmost parent will have run gc again and
922 ## hard-linked yet another old pack down to its children (not to mention
923 ## loose objects).
925 ## 2. When using the "-A" option with "git repack", any new objects in the
926 ## parent that are not referenced by children will continually get
927 ## exploded out of the hard-linked pack in the children whenever the
928 ## children run gc.
930 ## 3. To avoid suboptimal and/or unnecessarily many packs being hard-linked
931 ## into child forks, we must run the "mini" gc maintenance before we
932 ## perform the hard-linking into the children which provides yet another
933 ## source of inefficiency.
935 ## While we were still using the "-A" option to "git repack" (that was not
936 ## always the case) to guarantee we can access old ref values for long enough
937 ## to send out a meaningful mail.sh notification, another, more efficient,
938 ## option became available to prevent corruption of child forks that continue
939 ## to refer to objects that are no longer reachable from any ref in the parent.
941 ## The only things that need be copied (or hard-linked) into the child fork(s)
942 ## are those objects that have become unreachable from any ref in the parent.
944 ## When we were using the "git repack -A -d" + "git prune --expire=1.day.ago"
945 ## technique, the only objects that could ever be removed were loose objects
946 ## that "git prune" determined were expired. In that case, loose objects were
947 ## all that need be hard-linked down to child forks in order to avoid
948 ## corruption of any child fork(s).
950 ## The "git repack -A -d" + "git prune --expire=1.day.ago" + hard-linking loose
951 ## objects to child forks technique remains fundamentally sound from the
952 ## perspective of supporting simultaneous gc and push and keeping newly
953 ## unreachable objects around long enough to be sure we can send out meaningful
954 ## ref change notifications and never corrupting any child forks and never
955 ## persisting the lifetime of large old packs containing mostly duplicate or
956 ## unreachable objects as gc percolates through a project's entire fork tree.
958 ## However, that technique suffers from one potential prodigious pitfall.
960 ## Unreachable objects come flying out of their packs to splatter all over the
961 ## objects subdirectories possibly creating a huge, inefficient mess.
963 ## Often this is not an issue. Even with a lot of rebasing going on, usually
964 ## the only objects that will splatter are some commits, trees and the odd blob
965 ## here and there. Not enough to be overly concerned about.
967 ## However, for the reppository that frequently experiences a lot of non-fast-
968 ## forward updates and/or outright ref deletion, the number of objects suddenly
969 ## popping out of their packs at "git repack -A -d" time can be overwhelming.
971 ## To avoid this issue we now use a four phase pack creation strategy.
972 ## This will result in creation of up to four packs (instead of at most one).
974 ## I. A complete pack (with bitmaps if appropriate) gets created including
975 ## only "reachable" objects from all refs/... refs plus HEAD. This will
976 ## also serve as the virtual bundle for the repository.
978 ## II. A pack of recently-became-unreachable objects and friends is created.
979 ## (The "friends" are ref logs, linked working tree HEADs and indicies.)
980 ## Because both the pre-receive and update.sh script record all ref
981 ## changes we can easily choose the cut off point for "recently".
982 ## It is only the fact we maintain those logs in the reflogs subdirectory
983 ## that allows this step to be possible.
985 ## III. If the repository has any forks with a non-zero length alternates file,
986 ## yet another pack of "--keep-unreachable" objects is generated that will
987 ## not actually be kept in the parent, but hard-linked into all the forks.
989 ## IV. Finally, after running "git prune-packed", any remaining loose objects
990 ## are migrated into a pack of their own.
992 ## We then remove any non-.keep packs that existed before we started the
993 ## process being careful to keep any same-pack pushes for the "Push Pack Redux"
994 ## race condition (see docs/technical/gc.txt).
996 ## By using "git pack-objects" directly we are able to accomplish this with
997 ## very little additional effort.
999 ## The packs produced by (III) are treated almost like ".keep" packs by child
1000 ## forks in that the objects in them are never repacked into any other
1001 ## "--keep-unreachable" packs (but they can migrate into phase I or II packs)
1002 ## and those phase III packs are then hard-linked into any grandchild forks.
1004 ## This avoids the space explosion that could occur if each fork level ended
1005 ## up duplicating the "--keep-unreachable" pack space by repacking those
1006 ## objects (essentially breaking the hard-link to the single copy of those
1007 ## objects).
1009 ## While it is true that each level of forks could potentially add yet another
1010 ## phase III pack to be hard-linked down to its children, such packs will only
1011 ## include unreachable objects not already in any phase III packs that were
1012 ## received from the parent.
1014 ## The space for the phase III packs will not be reclaimed until the gc
1015 ## finishes percolating through the entire "fork tree" of a project.
1017 ## This is not much different than the "git repack -A -d" situation where
1018 ## all the loose objects are hard-linked down into child forks. In that
1019 ## case forks that actually need any of those objects could gradually reduce
1020 ## the number of objects hard-linked into deeper fork levels.
1022 ## The difference with a phase III "--keep-unreachable" pack is that there
1023 ## cannot be any gradual reduction like that since it would require repacking
1024 ## the pack and breaking the hard-link thereby increasing storage space. The
1025 ## storage will instead always be reclaimed all at once when all of the
1026 ## projects in the "fork tree" complete their gc.
1028 ## However, the belief is that the huge space win by having all the
1029 ## unreachable objects packed up together far eclipses (when many objects are
1030 ## involved, the single-pack version can end up using 1/20th or less of the
1031 ## disk space compared to having them all as loose objects) any brief minor
1032 ## space savings that might occur under the "git repack -A -d" loose object
1033 ## system prior to the gc collection completing for all the projects in the
1034 ## "fork tree".
1038 ## utility functions
1041 make_packs_ugw() {
1042 find -L "$1" -maxdepth 1 -type f ! -perm -ug+w \
1043 -name "pack-$octet20*.pack" -exec chmod ug+w '{}' + || :
1044 } 2>/dev/null
1046 get_index_tree() {
1047 if [ -s "$1" ]; then
1048 GIT_INDEX_FILE="$1"
1049 export GIT_INDEX_FILE
1050 git write-tree 2>/dev/null || :
1051 unset GIT_INDEX_FILE
1055 get_detached_head() {
1056 if [ -s "$1" ] && read -r _head <"$1" 2>/dev/null; then
1057 case "$_head" in $octet20*)
1058 echo "$_head"
1059 esac
1063 # single argument must be a top-level GIT_DIR
1064 # output will be (one per line, zero or more lines) full hash
1065 # values (i.e. 40 or more hex digits) from possible detached head
1066 # "single level" refs (e.g. "FETCH_HEAD" "MERGE_HEAD" etc.) from
1067 # files in the specified directory that have a "suitable" name and
1068 # have a length >= 40 and <= 1000 where EVERY line in a file MUST
1069 # match a 40 digit (or longer) hex string or that file will be ignored.
1070 # In a nod to Git each line first has any tab and anything following truncated.
1071 # NO sorting NOR "uniq"ing NOR "--batch-check"ing is performed on the output.
1072 # Also note that "HEAD" is NOT EXCLUDED and if detached will be output!
1073 get_detached_friends() {
1074 if [ -n "$1" ] && [ -d "$1" ]; then
1075 LC_ALL=C find -H "$1" -maxdepth 1 -name '[A-Za-z]*[A-Za-z0-9]' \
1076 -type f -size +39c -size -1001c -exec awk '
1077 BEGIN {exit}
1078 function process(fn, fnt, hashes, hcnt, fl, hi) {
1079 fnt = fn; sub(/^.*\//, "", fnt)
1080 if (length(fnt) > 31 || fnt !~ /^[A-Za-z][A-Za-z0-9_-]*[A-Za-z0-9]$/) return;
1081 hcnt = 0;
1082 while (getline fl < fn) {
1083 sub(/\t.*$/, "", fl)
1084 if (length(fl) < 40 || fl !~ /^[0-9a-fA-F][0-9a-fA-F]*$/) {hcnt = 0; break;}
1085 hashes[++hcnt] = fl;
1087 close(fn)
1088 for (hi=1;hi<=hcnt;++hi) print tolower(hashes[hi]);
1090 END {for (idx=1;idx<ARGC;++idx) process(ARGV[idx])}
1091 ' '{}' '+' 2>/dev/null || :
1095 # get_worktrees_friends
1096 # single argument is a (possibly relative) path to a "worktrees" subdir
1097 # if omitted it defaults to "worktrees"
1098 # any "friends" found in there are output to stdout
1099 get_worktrees_friends() {
1100 if [ -d "${1:-worktrees}" ]; then
1101 find -L "${1:-worktrees}" -mindepth 2 -maxdepth 2 -name HEAD -type f -print |
1102 while read -r _lwth; do
1103 get_detached_head "$_lwth"
1104 get_detached_friends "${_lwth%HEAD}"
1105 get_index_tree "${_lwth%HEAD}index"
1106 done
1110 # compute_extra_reachables
1111 # create lines suitable for a packed-refs file mentioning all the
1112 # other refs we might like to keep.
1113 # the current directory MUST be set to the repository's --git-dir
1114 # the following are included:
1115 # * refs mentioned in repack/packed-refs.extra (if it exists)
1116 # * refs mentioned in reflogs/... files
1117 # * tree(s) created from index file(s)
1118 # * detached linked working tree heads
1119 # Resulting objects are tested for existence and uniqified then output
1120 # one per line under a refs/z* namespace
1121 compute_extra_reachables() {
1123 if [ -s repack/packed-refs.extra ]; then
1124 LC_ALL=C sed <repack/packed-refs.extra -n \
1125 -e 's/^\([0-9A-Fa-f][0-9A-Fa-f]*\).*$/\1/p' \
1126 -e 's/^\^\([0-9A-Fa-f][0-9A-Fa-f]*\).*$/\1/p'
1128 digits8='[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]'
1129 find -L reflogs -mindepth 1 -maxdepth 1 -type f -name "$digits8*" -exec gzip -c -d -f '{}' + |
1130 LC_ALL=C awk '{print $2; print $3}'
1131 get_detached_friends .
1132 ! [ -f index ] || get_index_tree index
1133 get_worktrees_friends
1135 is_git_dir private &&
1136 [ "$(cd objects && pwd -P)" = "$(cd private/objects && pwd -P)" ] &&
1137 [ "$(cd refs && pwd -P)" != "$(cd private/refs && pwd -P)" ]
1138 then
1139 git --git-dir=private show-ref --head --hash 2>/dev/null || :
1140 get_detached_friends private
1141 ! [ -f private/index ] || get_index_tree private/index
1142 get_worktrees_friends private/worktrees
1144 } | LC_ALL=C sort -u |
1145 git cat-file ${var_have_git_260:+--buffer} --batch-check"${var_have_git_185:+=%(objectname)}" |
1146 LC_ALL=C awk '!/missing/ {num++; print $1 " " "refs/" substr("zzzzzzzzzzzz", 1, length(num)) "/" num}'
1150 ## main gc logic
1153 # Everything else is more efficient if we do this first
1154 # The "--prune" option is the default since v1.5.0 but it serves as "documentation" here
1155 git pack-refs --all --prune
1156 [ -e packed-refs ] || >>packed-refs # should never happen...
1158 # If we have a logs directory or a worktrees directory expire the ref logs now
1159 # Note that Git itself does not use either --rewrite or --updateref, so neither do we
1160 ! [ -d logs ] && ! [ -d worktrees ] || eval git reflog expire --all "${quiet:+>/dev/null 2>&1}" || :
1162 make_repack_dir
1163 ! [ -e .gc_failed ] || exit 1
1164 rm -f .gc_in_progress # make sure
1165 touch .gc_in_progress # it's truly fresh
1166 rm -f bundles/* objects/pack/pack-*.bndl
1167 # These only exist for a brief time before the packs loose their _f suffix
1168 # "Push Pack Redux" does not apply to these since they were only ever present with _f
1169 rm -f objects/pack/pack-*_f.keep
1170 # This is perhaps a bit aggressive in that if we're suffering from "Push Pack Redux"
1171 # and somehow we get run again immediately after the run where "Push Pack Redux" happened
1172 # and we have garbage collection forced, there's just the barest, almost negligible,
1173 # possibility that the "Push Pack Redux" ref updates _still_ have not happened and we
1174 # should not be removing _r .keep files. None of the normal Girocco processing can
1175 # cause this. The second run of this script would have to use the force gc option
1176 # for it to even be possible in the first place. What's much more likely is that
1177 # the initial run of this script was somehow interrupted in the middle before it
1178 # could get rid of the _r .keep file itself in which case it's better to get rid of
1179 # it now to avoid keeping something around that would perturb our nice and neat gc
1180 rm -f objects/pack/pack-*_r.keep
1181 # We will add .keep files for _u and _l packs if and when we run phase III
1182 # Otherwise they need to not have any .keep files during phases I and II
1183 rm -f objects/pack/pack-*_[ul].keep
1185 # We need to make sure that any non-Girocco (barely tolerated) Git object creation
1186 # activity will be able to "freshen" the pack containing a pre-existing object
1187 # that's being written. This really should not be necessary as the pre-receive
1188 # hook should make sure this takes place for any incoming pushes.
1189 # However, do it here anyway just in case.
1190 make_packs_ugw objects/pack
1192 # This is only effective with Git v2.3.5 and later and it will only matter when
1193 # we are using one of the "internal_rev_list" modes of pack-objects
1194 # (the combine-packs.sh script never uses any of those modes)
1195 # The "git repack" and "git prune" commands always set this internally themselves
1196 # It makes no difference if there's no repository corruption
1197 GIT_REF_PARANOIA=1 && export GIT_REF_PARANOIA
1199 # All of the options we might want to use with pack-objects were supported
1200 # at some point prior to Git version v1.6.6 which is the minimum version that
1201 # Girocco now requires. Except for one (--use-bitmap-index). Several of them
1202 # are "boiler plate" options we always want to use so we bundle them up here.
1203 pkopt="--delta-base-offset --keep-true-parents --non-empty --all-progress-implied"
1204 # We want to use --include-tag, but before Git v2.10.1 it would leave out
1205 # "middle" tags (e.g. a tag of a tag of a commit would omit the tagged tag)
1206 # See http://repo.or.cz/git.git/b773ddea2cd3b08c for details
1207 # ("pack-objects: walk tag chains for --include-tag", 2016-09-07, v2.10.1)
1208 # This is not a free check as it matches all refs against refs/tags/ then
1209 # peels all the annotated tags and checks for inclusion. The situation in
1210 # which it would add a tag that was not already included by a reachability
1211 # trace that included tag starting points can only occur if a new tag gets
1212 # pushed during gc pointing to something that would have been packed anyway.
1213 # But, it could happen and, really, compared to gc as a whole it's not that
1214 # expensive to perform (provided we do not get an unconnected pack).
1215 [ -z "$var_have_git_2101" ] || pkopt="$pkopt --include-tag"
1216 pkopt="$pkopt ${quiet:---progress} $packopts"
1218 # The git pack-objects command only supports bitmaps if all objects are being
1219 # packed (the "--all" option) and the "--stdout" option is NOT being used.
1220 # Additionally, while packing, if any encountered reachable objects are
1221 # determined to be "not wanted" then no bitmap index will be written anyway.
1222 # While it is theoretically possible that a project with a non-empty alternates
1223 # file ends up packing all objects (because it does not actually use any of the
1224 # objects found in the alternates), it's very unlikely. And, in the unlikely
1225 # event that did occur, clients would see a message about only using one bitmap
1226 # because Git can only use one bitmap at a time and at least one of the
1227 # alternates is bound to have a bitmap. Therefore if we see a non-empty
1228 # alternates file, we disable writing bitmaps which avoids the warning and any
1229 # possibility of a client warning as well. Also if we are running anything
1230 # before Git v2.1.0 (the effective version for repack.writeBitmaps=true) then
1231 # we also always disable bitmap writing.
1232 wbmopt=
1233 [ -z "$var_have_git_210" ] || wbmopt="--write-bitmap-index"
1234 # More recent versions of pack-objects have optimizations when not using the
1235 # --local option. If we do not have any alternates it's a pointless option.
1236 # If we do have alternates we need to skip writing a bitmap and we cannot
1237 # have a bundle since it must contain all objects.
1238 if [ -n "$isfork" ]; then
1239 lclopt="--local"
1240 wbmopt=
1241 makebndl=
1242 else
1243 lclopt=
1244 makebndl=1
1248 ## Phase I
1251 wbmstr=
1252 [ -n "$wbmopt" ] || wbmstr=" (bitmaps disabled)"
1253 progress "~ [$proj] running primary full gc pack-objects$wbmstr ($(date))"
1255 gotforks=
1256 ! has_forks_with_alternates "$proj" || gotforks=1
1258 # To avoid "Push Pack Redux" (see docs/technical/gc.txt), after collecting the
1259 # initial preexisting non-keep pack list, we rename them so that an incoming push
1260 # pack cannot possibly experience a pack name collision. Git does not require
1261 # use of the "default" pack names, simply that the proper extensions are used.
1262 # We rename to insert an "_r" just before the extension to avoid "Push Pack Redux"
1263 # name collisions. Later on we may create an "unreachable" pack for hard-linking
1264 # down into forks and it will have an "_u" inserted just before its extension.
1265 packlist="$(list_packs -C objects/pack --all --exclude-no-idx --exclude-keep --quiet .)" || :
1266 oldpacks=
1267 for oldpack in $packlist; do
1268 oldpack="${oldpack%.pack}"
1269 [ -f "objects/pack/$oldpack.pack" ] || {
1270 echo >&2 "[$proj] unable to list old pack files"
1271 exit 1
1273 case "$oldpack" in pre-auto-gc-[12])
1274 # we never disturb pre-auto-gc-1 or pre-auto-gc-2 packs
1275 continue
1276 esac
1277 oldpackhex="${oldpack#pack-}"
1278 if [ "${oldpackhex#*[!0-9a-fA-F]}" != "$oldpackhex" ]; then
1279 # names not exclusively hexadecimal do not need renaming
1280 case "$oldpack" in
1281 pack-$octet20*_l)
1282 # _l packs are treated like still-unpacked loose objects
1283 continue;;
1284 *_f)
1285 # _f packs can only be left over from a previously interrupted gc;
1286 # they need to be renamed to _r now so they're not confused with
1287 # any freshly generated "final" packs (and we already removed
1288 # any pre-existing *_f.keep files so we're good to go)
1291 oldpacks="${oldpacks:+$oldpacks }$oldpack"
1292 continue;;
1293 esac
1295 rename_pack "objects/pack/$oldpack" "objects/pack/${oldpack%_f}_r" || {
1296 echo >&2 "[$proj] unable to rename old pack files"
1297 exit 1
1299 # If the oldpack has a .keep now it means a "Push Pack Redux" is actually
1300 # in progress at this moment and we need to .keep the renamed pack,
1301 # otherwise no "Push Pack Redux" has started yet or it has already finished.
1302 # In either case we're okay because if it's just finished then all ref
1303 # changes have already been made so we don't need a .keep and we will
1304 # see the ref changes and grab all the objects via a reachability trace.
1305 # If it hasn't started yet that's okay because we're done moving that
1306 # name so a complete pack will appear under the old name that we'll
1307 # leave alone.
1308 if [ -f "objects/pack/$oldpack.keep" ]; then
1309 echo "Push Pack Redux" >"objects/pack/${oldpack%_f}_r.keep"
1310 else
1311 oldpacks="${oldpacks:+$oldpacks }${oldpack%_f}_r"
1313 done
1315 # We wish to keep deltas from our last full pack so if we're not redeltaing
1316 # then make sure the .pack associated with the .bitmap has a newer mod time
1317 # (If there is no .bitmap then touch the pack with the most objects instead.)
1318 if [ -z "$newdeltas" ]; then
1319 bmpack="$(list_packs --exclude-no-bitmap --exclude-no-idx --max-matches 1 objects/pack)"
1320 [ -n "$bmpack" ] || bmpack="$(list_packs --exclude-no-idx --max-matches 1 --object-limit -1 --include-boundary objects/pack)"
1321 if [ -n "$bmpack" ] && [ -f "$bmpack" ] && [ -s "$bmpack" ]; then
1322 sleep 1
1323 touch -c "$bmpack" 2>/dev/null || :
1324 # We must touch .gc_in_progress here to avoid $bmpack looking
1325 # like it's been "freshened" when redundant packs are removed
1326 # It's okay if they have the same mod time, but POSIX does not
1327 # guarantee an ordering for the "touching" that occurs which is
1328 # why this must be a separate command but needs no "sleep 1"
1329 touch .gc_in_progress
1333 # Now we need to make sure that any "freshening" that takes place will actually
1334 # result in a "newer" modification time than the .gc_in_progress file now has
1335 sleep 1
1337 # We run git pack-objects from the repack subdirectory so we can force
1338 # optimized packs to be generated even for repositories that do not have any
1339 # tagged commits
1340 packs="$(git --git-dir=repack pack-objects </dev/null \
1341 $pkopt --all $newdeltas $lclopt ${wbmopt:---honor-pack-keep} "$@" repack/alt/pack/pack)"
1342 v_cnt packcnt $packs
1343 [ $packcnt -eq 1 ] || makebndl=
1346 ## Phase II
1349 progress "~ [$proj] running supplementary gc pack-objects ($(date))"
1351 # Add the "supplementary" refs
1352 compute_extra_reachables >>repack/packed-refs
1354 # Subtract the primary refs
1355 GIT_ALTERNATE_OBJECT_DIRECTORIES="$PWD/repack/alt"
1356 export GIT_ALTERNATE_OBJECT_DIRECTORIES
1358 # For this one we MUST use --local and MUST NOT use --write-bitmap-index
1359 # However, if there is a "logs" subdirectory we need to use --reflog
1360 # We do add it, just in case, if the linked working trees dir is present
1361 # We do not add --indexed-objects as that requires v2.2.0 and it's unclear
1362 # if it properly includes linked working tree index files or not. The
1363 # above compute_extra_reachables has already included all index trees (thereby
1364 # providing proper --indexed-objects support for all Git versions) making the
1365 # option completely unnecessary.
1366 rflopt=
1367 ! [ -d logs ] && ! [ -d worktrees ] || rflopt=--reflog
1368 spacks="$(git --git-dir=repack pack-objects </dev/null \
1369 $pkopt --honor-pack-keep --all $rflopt $newdeltas --local "$@" repack/alt/pack/pack)"
1372 ## Phase III
1375 # There's nothing to do for Phase III unless we have forks that refer to our
1376 # project from their alternates file
1377 hlpacks=
1378 upacks=
1379 if [ -n "$gotforks" ]; then
1381 progress "~ [$proj] running keep-unreachable gc pack-objects for forks ($(date))"
1383 # If we are a fork, any pre-existing _u packs need to have a .keep
1384 # for this phase and be added to the hlpacks list otherwise (we are
1385 # not a fork) pre-existing _u packs are anomalies to be treated like
1386 # regular non-_u packs
1387 if [ -n "$isfork" ]; then
1388 for upack in $(find -L objects/pack -mindepth 1 -maxdepth 1 -name "pack-$octet20*_[ul].pack" -print); do
1389 upack="${upack%.pack}"
1390 [ -e "$upack.keep" ] || echo "unreachable" >"$upack.keep"
1391 case "$upack" in *_l);;*)
1392 hlpacks="${hlpacks:+$hlpacks }${upack#objects/pack/pack-}"
1393 esac
1394 done
1396 # Using either --no-reuse-delta or --no-reuse-object together with the
1397 # --keep-unreachable option is a very, very, very bad idea when good
1398 # packs are the desired outcome. If newdeltas are being generated
1399 # then we pack to a temp name, and use combine-packs.sh to get a better
1400 # pack as the result to avoid making a bad --keep-unreachable pack
1401 pfx=
1402 [ -z "$newdeltas" ] || pfx="ku"
1403 upacks="$(git --git-dir=repack pack-objects </dev/null \
1404 $pkopt --honor-pack-keep --all $rflopt --keep-unreachable --local "$@" repack/alt/pack/${pfx}pack)"
1405 if [ -n "$upacks" ] && [ -n "$newdeltas" ]; then
1406 progress "~ [$proj] rebuilding keep-unreachable pack deltas"
1407 oldupacks="$upacks"
1408 upacks="$(
1409 printf "repack/alt/pack/${pfx}pack-%s.pack\n" $oldupacks |
1410 run_combine_packs --names --weak-naming --non-empty --all-progress-implied ${quiet:---progress} \
1411 $packopts $newdeltas "$@" repack/alt/pack/pack)"
1412 eval rm -f "$(printf \""repack/alt/pack/${pfx}pack-%s.*"\"" " $oldupacks)"
1414 for upack in $upacks; do
1415 rename_pack "repack/alt/pack/pack-$upack" "repack/alt/pack/pack-${upack}_u"
1416 done
1417 rm -f objects/pack/pack-*_[ul].keep
1418 [ -z "$hlpacks" ] && [ -z "$upacks" ] ||
1419 progress "~ [$proj] hard-linking keep-unreachable pack(s) into immediate child forks"
1421 # We have to update the lastparentgc time in the child forks even if they do not get any
1422 # new "unreachable packs" because they need to run gc just in case the parent now has some
1423 # objects that used to only be in the child so they can be removed from the child.
1424 # For example, a "patch" might be developed first in a fork and then later accepted into
1425 # the parent in which case the objects making up the patch in the child fork are now
1426 # redundant (since they're now in the parent as well) and need to be removed from the
1427 # child fork which can only happen if the child fork runs gc.
1428 lastparentgc="$(date "$datefmt")"
1430 # It is enough to copy objects just one level down and get_repo_list
1431 # takes a regular expression (which is automatically prefixed with '^')
1432 # so we can easily match forks exactly one level down from this project
1433 forkdir="$proj"
1434 get_repo_list "$forkdir/[^/:][^/:]*:" |
1435 while read fork; do
1436 # Ignore forks that do not exist or are symbolic links
1437 ! [ -L "$cfg_reporoot/$fork.git" ] && [ -d "$cfg_reporoot/$fork.git" ] ||
1438 continue
1439 # Or have an empty alternates file
1440 ! is_empty_alternates_file "$cfg_reporoot/$fork.git/objects/info/alternates" ||
1441 continue
1442 runupdate=
1443 # Match hlpacks in parent project if any
1444 if [ -n "$hlpacks" ]; then
1445 mkdir -p "$cfg_reporoot/$fork.git/objects/pack"
1446 eval ln -f "$(printf '"objects/pack/pack-%s.pack" ' $hlpacks)" \
1447 "$(printf '"objects/pack/pack-%s.idx" ' $hlpacks)" \
1448 '"$cfg_reporoot/$fork.git/objects/pack/"'
1449 runupdate=1
1451 # Match upacks in repack/alt area if any
1452 if [ -n "$upacks" ]; then
1453 mkdir -p "$cfg_reporoot/$fork.git/objects/pack"
1454 eval ln -f "$(printf '"repack/alt/pack/pack-%s_u.pack" ' $upacks)" \
1455 "$(printf '"repack/alt/pack/pack-%s_u.idx" ' $upacks)" \
1456 '"$cfg_reporoot/$fork.git/objects/pack/"'
1457 runupdate=1
1459 if ! [ -e "$cfg_reporoot/$fork.git/.needsgc" ]; then
1460 # Trigger a mini gc in the fork if it now has too many packs
1461 packs="$(list_packs --quiet --count --exclude-no-idx --exclude-keep "$cfg_reporoot/$fork.git/objects/pack")" || :
1462 if [ -n "$packs" ] && [ "$packs" -ge 20 ]; then
1463 >"$cfg_reporoot/$fork.git/.needsgc"
1466 [ -z "$runupdate" ] || git --git-dir="$cfg_reporoot/$fork.git" update-server-info
1467 # Update the fork's lastparentgc date (must be more recent than $gcstart)
1468 git --git-dir="$cfg_reporoot/$fork.git" config gitweb.lastparentgc "$lastparentgc"
1469 done
1472 # Now move any primary/supplementary packs back into objects/pack
1473 # then drop any "unfreshened" redundant packs and clear repack/alt
1475 # First make sure the primary pack(s) have the most recent mod time
1476 if [ -n "$packs" ]; then
1477 [ -z "$spacks" ] || sleep 1
1478 printf 'repack/alt/pack/pack-%s.pack\n' $packs | xargs touch -c 2>/dev/null || :
1481 # Move the packs into place but with a _f suffix and a .keep file for now
1482 for pack in $packs $spacks; do
1483 rename_pack "repack/alt/pack/pack-$pack" "objects/pack/pack-${pack}_f"
1484 [ -e "objects/pack/pack-${pack}_f.keep" ] ||
1485 echo "final" >"objects/pack/pack-${pack}_f.keep"
1486 done
1488 # It's possible that one of the $oldpacks had a .bitmap, got renamed (along
1489 # with its .bitmap) and then got "freshened" causing us to not remove it
1490 # However, if $wbmopt is set we most likely now have TWO .bitmap packs!
1491 # This can produce ugly warnings we don't want and possibly get the wrong
1492 # bitmap used since only one .bitmap file can ever be used by Git.
1493 # If this has happened, the .bitmap we want to discard will always have
1494 # an _r suffix so we can just zap any such now since it will leave the pack.
1495 [ -z "$wbmopt" ] || rm -f objects/pack/pack-*_r.bitmap || :
1497 # Remove the redundant packs that have not since been "freshened"
1498 # This does not completely eliminate the race condition window (Girocco's own
1499 # activites -- gc/fetch/receive are immune to the race) but it substantially
1500 # shrinks it down to just the time after the find but before the following rm
1501 >repack/oldpacks
1502 [ -z "$oldpacks" ] ||
1503 printf 'objects/pack/%s.pack\n' $oldpacks |
1504 LC_ALL=C sort >repack/oldpacks
1505 find -L objects/pack -maxdepth 1 -type f -name "pack-$octet20*.pack" -newer .gc_in_progress -print |
1506 LC_ALL=C sort >repack/freshened
1507 deadpacks="$(LC_ALL=C join -v 1 repack/oldpacks repack/freshened | LC_ALL=C sed 's/\.pack$//')"
1508 [ -z "$deadpacks" ] ||
1509 eval echo "$(printf '"%s".* ' $deadpacks)" | xargs rm -f || :
1511 # No need for this anymore
1512 rm -rf repack/alt objects/pack/repack
1513 unset GIT_ALTERNATE_OBJECT_DIRECTORIES
1516 ## Phase IV
1519 progress "~ [$proj] running gc prune-packed"
1521 # We do not want the redundant packs or any new "--keep-unreachable" pack(s) to be
1522 # present while running prune-packed. We try to guarantee that any loose object
1523 # (or any object present in a pack with an _l suffix which was created by mini gc)
1524 # that's unreachable persists for at least one $Girocco::Config::min_gc_interval
1525 # (not withstanding administrator interference to force earlier gc to occur).
1526 # If we were to include the redundant/keep-unreachable pack(s) when running
1527 # prune-packed and a loose unreachable object happened to be duplicated in one
1528 # of them we would end up removing it too soon and void our guarantee.
1529 git prune-packed $quiet
1531 progress "~ [$proj] running loose objects gc pack-objects ($(date))"
1533 # Although Git v2.10.0 and later support a --pack-loose-unreachable option,
1534 # we MUST NOT use it for these reasons:
1535 # 1) We're not interested in expensive "unreachable" at this point, only "loose"
1536 # 2) It produces simply horrid packs about 3.8x times larger than they should be
1537 # 3) We don't require anything more than Git v1.6.6
1538 # The only way we could see any _o pack files at this point is if one got
1539 # "freshened" while we were running gc. If that happens then it gets to live on
1540 # until the next full gc and we need to include it in the loose repack here.
1541 lpacks="$(list_packs --exclude-no-idx --exclude-no-sfx _l --exclude-no-sfx _o --quiet objects/pack |
1542 run_combine_packs --replace --names --loose --weak-naming --non-empty --honor-pack-keep \
1543 --all-progress-implied ${quiet:---progress} $packopts $newdeltas "$@")"
1545 if [ -n "$lpacks" ]; then
1546 # Make sure any primary pack(s) have a more recent mod time than "unreachable" objects packs
1547 if [ -n "$packs" ]; then
1548 sleep 1
1549 printf 'objects/pack/pack-%s_f.pack\n' $packs | xargs touch -c 2>/dev/null || :
1551 # We need to identify these packs later so we don't combine_packs them
1552 for objpack in $lpacks; do
1553 rename_pack "objects/pack/pack-$objpack" "objects/pack/pack-${objpack}_o" || :
1554 done
1557 # Polish up the final packs now
1558 rm -f objects/pack/pack-*_f.keep
1559 for pack in $packs $spacks; do
1560 rename_pack "objects/pack/pack-${pack}_f" "objects/pack/pack-$pack"
1561 done
1563 if [ -n "$lpacks" ]; then
1564 # Finally zap the corresponding loose objects
1565 progress "~ [$proj] running packed loose objects gc prune-packed"
1566 git prune-packed $quiet
1569 ! [ -e .gc_failed ] || exit 1
1570 # These, if they exist, are now meaningless and need to be removed
1571 rm -f gfi-packs .needsgc .needspack .needspackgc
1573 # Make sure this stays up to date
1574 git update-server-info
1576 # We must make loose objects group writable so that they
1577 # can be freshened by other pushers. Technically we need only do this for
1578 # push projects but to enable mirror projects to be more easily converted to
1579 # push projects, we go ahead and do it for all projects.
1580 # By the time we get here we really shouldn't have any of these, but just in case.
1581 { find -L objects/$octet -type f -name "$octet19*" -exec chmod ug+w '{}' + || :; } 2>/dev/null
1583 # darcs mirrors have a xxx.log file that will grow endlessly
1584 # if this is a mirror and the file exists, shorten it to 10000 lines
1585 # also take this opportunity to optimize the darcs repo
1586 if ! [ -e .nofetch ] && [ -n "$cfg_mirror" ]; then
1587 url="$(config_get baseurl)" || :
1588 case "$url" in darcs://* | darcs+http://* | darcs+https://*)
1589 if [ -n "$cfg_mirror_darcs" ]; then
1590 url="${url%/}"
1591 basedarcs="$(basename "${url#darcs*:/}")"
1592 if [ -f "$basedarcs.log" ]; then
1593 tail -n 10000 "$basedarcs.log" >"$basedarcs.log.$$"
1594 mv -f "$basedarcs.log.$$" "$basedarcs.log"
1596 if [ -d "$basedarcs.darcs" ]; then
1598 cd "$basedarcs.darcs"
1599 # without show_progress suppress non-error output
1600 [ "${show_progress:-0}" != "0" ] || exec >/dev/null
1601 # Note that this does not optimize _darcs/inventories/ :(
1602 darcs optimize || :
1606 esac
1609 # Create a matching .bndl header file for the all-in-one pack we just created
1610 # but only if we're not a fork (otherwise the bundle would not be complete)
1611 # and we are running at least Git version 1.7.2 (pack_is_complete always fails otherwise)
1612 if [ -n "$makebndl" ] && [ -n "$var_have_git_172" ]; then
1613 # There should only be one pack in $packs but do some checking...
1614 # The one we just created will have a .idx and will NOT have a .keep
1615 progress "~ [$proj] creating downloadable bundle header"
1616 pkbase=
1617 pkhead=
1618 IFS= read -r curhead <repack/HEAD.orig || :
1620 [ -s "objects/pack/pack-$packs.pack" ] &&
1621 [ -s "objects/pack/pack-$packs.idx" ] &&
1622 ! [ -e "objects/pack/pack-$packs.keep" ] &&
1623 pkhead="$(pack_is_complete "$PWD/objects/pack/pack-$packs.pack" \
1624 "$PWD/repack/packed-refs.orig" "$curhead")"
1625 then
1626 pkbase="objects/pack/pack-$packs"
1628 if [ -n "$pkbase" ] && [ -n "$pkhead" ]; then
1630 symref=
1631 case "$curhead" in "ref: refs/"?*|"ref:refs/"?*|"refs/"?*)
1632 symref="${curhead#ref:}"
1633 symref="${symref# }"
1634 esac
1635 bndlurl=
1636 [ -z "$cfg_httpbundleurl" ] || bndlurl=" url=$cfg_httpbundleurl/$proj.git/clone.bundle"
1637 echo "# v2 git bundle"
1638 LC_ALL=C sed -ne "/^$octet20$hexdig* refs\/[^ $tab]*\$/ p" <repack/packed-refs.orig
1639 if [ -n "$symref" ]; then
1640 printf "$pkhead HEAD\0symref=HEAD:%s%s\n" "$symref" "$bndlurl"
1641 else
1642 if [ -n "$bndlurl" ]; then
1643 printf "$pkhead HEAD\0%s\n" "${bndlurl# }"
1644 else
1645 echo "$pkhead HEAD"
1648 echo ""
1649 } >"$pkbase.bndl"
1650 bndletag="$("$cfg_basedir/bin/rangecgi" --etag -m 1 "$pkbase.bndl" "$pkbase.pack")" || :
1651 bndlsha="$(printf '%s' "$bndletag" | git hash-object --stdin)" || :
1652 if [ -n "$bndletag" ]; then
1653 case "$bndlsha" in $octet20*)
1654 bndlshatrailer="${bndlsha#????????}"
1655 bndlshaprefix="${bndlsha%$bndlshatrailer}"
1656 bndlname="$(TZ=UTC date +%Y%m%d_%H%M%S)-${bndlshaprefix:-0}"
1657 [ -d bundles ] || mkdir bundles
1658 echo "${pkbase#objects/pack/}.bndl" >"bundles/$bndlname"
1659 echo "${pkbase#objects/pack/}.pack" >>"bundles/$bndlname"
1660 ln -s -f -n "$bndlname" bundles/latest
1661 esac
1666 # Record the size of this repo as the sum of its clone packed-refs + *.pack sizes as 1024-byte blocks
1667 eval "reposizek=$(( $(
1668 echo 0 $(du -k repack/packed-refs.orig $(printf 'objects/pack/pack-%s.pack ' $packs) 2>/dev/null |
1669 LC_ALL=C awk '{print $1}') |
1670 LC_ALL=C sed -e 's/ / + /g') ))"
1671 config_set_raw girocco.reposizek "${reposizek:-0}"
1673 # Now we're finally done with this
1674 rm -rf repack
1676 # We didn't used to do anything about rerere or worktrees but we're
1677 # trying to make nice with linked working trees these days :)
1678 # Maybe even non-bare repositories too, but *shush* about those ;)
1679 if [ -n "$var_have_git_250" ] && [ -d worktrees ]; then
1680 # The value "3.months.ago" is hard-coded into gc.c rather than
1681 # having the default be in worktree.c so we must provide it if
1682 # we get nothing out of the gc.worktreePruneExpire config item
1683 # Prior to Git v2.6.0 the config item was gc.pruneworktreesexpire
1684 # however we just always use the newer name no matter what Git version
1685 expiry="$(git config --get gc.worktreePruneExpire 2>/dev/null)" || :
1686 eval git worktree prune --expire '"${expiry:-3.months.ago}"' "${quiet:+>/dev/null 2>&1}" || :
1688 # git rerere does it right and handles its own default/config'd expiration values
1689 ! [ -d rr-cache ] || eval git rerere gc "${quiet:+>/dev/null 2>&1}" || :
1691 # We use $gcstart here to avoid a race where a push occurs during the gc itself
1692 # and the next future gc could be incorrectly skipped if we used the current
1693 # timestamp here instead
1694 config_set lastgc "$gcstart"
1695 rm -f "$lockf"
1697 progress "- [$proj] garbage check ($(date))"