jobd/gc.sh: compute simple gc "signature" hashes
[girocco.git] / jobd / gc.sh
blobac104bf74dc8a9d7b7560d1f1ab0d5532983546b
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 # Compute a "signature" for the refs in the specified (default .) repository.
334 # Instead of passing --head to show-ref, the actual contents of HEAD are used
335 # to capture changes in HEAD's symbolic-ref value that would otherwise be missed.
336 # $1 => variable name to store result hash in
337 # $2 => --git-dir to use, defaults to "."
338 v_compute_refs_sighash() {
339 eval "$1="'"$({ cat "${2:-.}/HEAD"; git --git-dir="${2:-.}" show-ref; } |
340 git hash-object -t blob --stdin)"'
343 # On return a "$lockf" will have been created that must be removed when gc is done
344 lock_gc() {
345 v_lock_gc _lockresult || {
346 echo >&2 "[$proj] $_lockresult"
347 exit 1
349 lockf="$_lockresult"
352 # Create a repack subdirectory such that running repack in it will pack the
353 # same things that a pack in the normal directory would except that the pack
354 # is guaranteed to be generated in an optimized order by adding a suitable
355 # synthesized ref in the refs/tags namespace (yes, pack-objects.c really does
356 # behave differently depending on the contents of the refs/tags namespace).
357 # Before calling this, pack-refs --all MUST be performed or the wrong pack
358 # will end up being made.
360 # If a ref deletion is pushed after making the repack subdir but before the
361 # the actual repack, the discarded objects will be packed -- no big deal,
362 # they'll get discarded the next time gc runs.
364 # If a fast-forward ref update is pushed after making the repack subdir but
365 # before the actual repack, it will be picked up and the new objects packed
366 # (subject to the normal git repack race about picking such updates up).
368 # If a non-fast-forward ref update is pushed after making the repack subdir but
369 # before the actual repack, it will be picked up like a fast-forward update but
370 # the discarded objects will be included like a ref deletion (until the next
371 # scheduled gc takes place).
373 # We retain a copy of the original packed-refs file as repack/packed-refs.orig
374 # If ref deletions come in while we're repacking, the original packed-refs
375 # file will be modified, but we'll still pack the deleted ref(s).
376 # If the packed-refs.orig file is used to create the bundle header we avoid
377 # a situation where the bundle contains a ref state that never actually
378 # existed in reality (for example a new branch is pushed and then an old
379 # branch deleted afterwards -- the deletion would show up in the bundle
380 # because it will cause the original packed-refs file to be re-written, but
381 # the new branch creation will not unless we do another pack-refs which might
382 # lead to having in incomplete bundle). Therefore we want to keep a copy of
383 # the original packed-refs file around. We do the same thing for HEAD.
385 # It's possible that the "objects" subdirectory is a symbolic link.
386 # Git does support this. However, during the repacking process, new packs
387 # will be created in repack/alt/pack and then moved into objects/pack.
388 # In order for this to work seemlessly, they must both be on the same
389 # filesystem. But when objects (or even objects/pack) is a symbolic link they
390 # might not be. For this reason a "repack" subdirectory is created under
391 # objects/pack and the repack/alt/pack directory symbolicly linked to it.
393 # Git allows not just HEAD to be a symbolic-ref, but any ref anywhere in the
394 # refs namespace. We are concerned about ref name collisions and getting the
395 # right tag set to get an optimal pack. We can safely duplicate the ref space
396 # under refs/heads, refs/notes and refs/remotes without any risk of unwanted
397 # collisions and this will likely make over 99%+ of all symbolic refs found
398 # in the wild work properly. Girocco itself never creates any symbolic refs
399 # inside the refs namespace; this is a nod to simultaneously using a Girocco
400 # repository for other purposes.
401 make_repack_dir() {
402 ! [ -d repack ] || rm -rf repack
403 ! [ -d repack ] || { echo >&2 "[$proj] cannot remove repack subdirectory"; exit 1; }
404 [ -d objects/pack ] || mkdir -p objects/pack
405 ! [ -d objects/pack/repack ] || rm -rf objects/pack/repack
406 ! [ -d objects/pack/repack ] || { echo >&2 "[$proj] cannot remove objects/pack/repack subdirectory"; exit 1; }
407 mkdir repack repack/refs repack/alt objects/pack/repack
408 [ -d info ] || mkdir info
409 ln -s ../config repack/config
410 ln -s ../info repack/info
411 ln -s ../objects repack/objects
412 ln -s "$PWD/objects/pack/repack" repack/alt/pack
413 ! [ -d logs ] || ln -s ../logs repack/logs
414 ! [ -d worktrees ] || ln -s ../worktrees repack/worktrees
415 _lines=$(( $(LC_ALL=C wc -l <packed-refs) ))
416 cat HEAD >repack/HEAD.orig
417 cat packed-refs >repack/packed-refs.orig
418 ln -s packed-refs.orig repack/packed-refs
419 cp repack/HEAD.orig repack/HEAD
420 v_compute_refs_sighash _rsig repack
421 rm repack/packed-refs repack/HEAD
422 ln -s ../../refs repack/refs/refs
423 echo "$_rsig" >repack/packed-refs.sighash
424 >repack/packed-refs.extra
425 _xtralines=0
426 if [ $(LC_ALL=C wc -l <repack/packed-refs.orig) -ne "$_lines" ]; then
427 echo >&2 "[$proj] error: make_repack_dir failed original packed-refs line count sanity check"
428 exit 1
430 if [ "${cfg_fetch_stash_refs:-0}" = "0" ]; then
431 # migrate any refs/stash or refs/tgstash lines to repack/packed-refs.extra
432 <repack/packed-refs.orig LC_ALL=C awk -v xtra="repack/packed-refs.extra" '
433 BEGIN { peeling = 0 }
434 NR == 1 && /^#/ { print; next; }
435 peeling && /^\^/ { print >>xtra; next; }
436 /^[0-9A-Fa-f][0-9A-Fa-f][0-9A-Fa-f][0-9A-Fa-f]+ refs\/(stash|tgstash)(\/|$)/ {
437 peeling = 1
438 print >>xtra
439 next
441 { peeling = 0; print; }
442 ' >repack/packed-refs.new
443 _xtralines="$(( $(LC_ALL=C wc -l <repack/packed-refs.extra) + 0 ))"
444 _newlines="$(( $(LC_ALL=C wc -l <repack/packed-refs.new) + 0 ))"
445 if [ "$(( $_newlines + $_xtralines ))" -ne "$_lines" ]; then
446 echo >&2 "[$proj] error: make_repack_dir failed packed-refs.extra line count sanity check"
447 exit 1
449 mv -f repack/packed-refs.new repack/packed-refs.orig
450 _lines="$_newlines"
452 # Note: Git v1.5.0 introduced the "# pack-refs with:" header line for the packed-refs file
453 sed '/^# pack-refs/d; s, refs/, refs/!/,' <repack/packed-refs.orig >repack/packed-refs
454 nohead=
455 headref="$(git rev-parse --verify --quiet HEAD)" || :
456 if [ -n "$headref" ]; then
457 echo "$headref refs/!=/HEAD" >>repack/packed-refs
458 echo "$headref refs/heads/!" >>repack/packed-refs
459 nohead='\, refs/heads/!$,d; '
460 _lines=$(( $_lines + 2 ))
462 if [ $(( $(LC_ALL=C wc -l <repack/packed-refs) + 1 )) -ne "$_lines" ]; then
463 echo >&2 "[$proj] error: make_repack_dir failed packed-refs initial line count sanity check"
464 exit 1
466 sed -n "$nohead"'\, refs/heads/,p; \, refs/notes/,p; \, refs/remotes/,p' <repack/packed-refs.orig >>repack/packed-refs
467 _newlines="$(( $(LC_ALL=C wc -l <repack/packed-refs) ))"
468 if [ $(( $_newlines + 1 )) -lt "$_lines" ]; then
469 echo >&2 "[$proj] error: make_repack_dir failed packed-refs extra line count sanity check"
470 exit 1
472 _lines="$_newlines"
473 optref="$(git rev-list -n 1 --all 2>/dev/null)" || :
474 if [ -n "$optref" ]; then
475 echo "$optref refs/tags/!" >>repack/packed-refs
476 _lines=$(( $_lines + 1 ))
477 echo "$optref" >repack/HEAD
478 else
479 cat HEAD >repack/HEAD
481 if [ $(LC_ALL=C wc -l <repack/packed-refs) -ne "$_lines" ]; then
482 echo >&2 "[$proj] error: make_repack_dir failed packed-refs line count sanity check"
483 exit 1
487 # Remove any crud that's been left behind by interrupted operations
488 # that did not clean up after themselves
489 remove_crud() {
490 # Remove any existing FETCH_HEAD
491 # There can only be a FETCH_HEAD if we've been fetching, not if we've been
492 # receiving pushes (those never create a FETCH_HEAD).
493 # And if we're fetching because we're a mirror, we know we're not fetching right
494 # now since jobd.pl never runs a project's fetch simultaneously with its gc.
495 # Therefore any existing FETCH_HEAD is junk. And it may be many megabytes if
496 # there were a lot of refs.
497 rm -f FETCH_HEAD
499 # Remove any stale ref locks
500 clear_stale_ref_locks
502 # remove any existing pack_is_complete_test or repack subdirectories
503 # If either exists when this function is called it's crud
504 rm -rf pack_is_complete_test repack objects/pack/repack
506 # Remove any stale pack remnants that are more than an hour old.
507 # Stale pack fragments are defined as any pack-<sha1>.ext where .ext is NOT
508 # .pack AND the corresponding .pack DOES NOT exist. A bunch of stale
509 # pack-<sha1>.idx files without their corresponding .pack files are worthless
510 # and just waste space. Normally there shouldn't be any remnants but actually
511 # this can happen when things are interrupted at just the wrong time.
512 # Note that the objects/pack directory is created by git init and should
513 # always exist.
514 find -L objects/pack -maxdepth 1 -type f -mmin +60 -name "pack-$octet20*.?*" -print |
515 LC_ALL=C sed -e 's/^objects\/pack\/pack-//; s/\..*$//' | LC_ALL=C sort -u |
516 while read packsha; do
517 ! [ -e "objects/pack/pack-$packsha.pack" ] || continue
518 rm -f "objects/pack/pack-$packsha".?*
519 done
521 # Remove any stale tmp reflogs files that are more than one hour old.
522 # Since they are created only while the pre-receive hook is running and
523 # all it does is process a bunch of refs passed to it on standard input
524 # it's inconceivable that it would ever take as much as an hour to run.
525 if [ -d reflogs ]; then
526 find -L reflogs -maxdepth 1 -type f -mmin +60 -name "tmp_*" -exec rm -f '{}' + || :
529 # Remove any stale object tmp_obj_* files that are more than 3 hours old.
530 # Really these files should only exist very briefly so there shouldn't be any
531 # but things happen that can end up leaving them behind.
532 find -L objects/$octet -maxdepth 1 -type f -mmin +180 -name "tmp_obj_?*" -exec rm -f '{}' + 2>/dev/null || :
534 # Remove any stale pack .keep files that are more than 12 hours old.
535 # We don't do anything to create any permanent pack .keep files, so they must
536 # be remnants from some failed push or something. Removing the .keep will
537 # allow the pack to be properly repacked.
538 find -L objects/pack -maxdepth 1 -type f -mmin +720 -name "pack-$octet20*.keep" -exec rm -f '{}' + || :
540 # Remove any stale tmp_pack_*, tmp_idx_*, tmp_bitmap_*, packtmp-* or .tmp-*-pack* files
541 # that are more than 12 hours old.
542 find -L objects/pack -maxdepth 1 -type f -mmin +720 \( \
543 -name "tmp_pack_?*" -o -name "tmp_idx_?*" -o -name "tmp_bitmap_?*" -o \
544 -name "packtmp-?*" -o -name ".tmp-?*-pack*" \
545 \) -exec rm -f '{}' + || :
547 # Remove any stale incoming-* object quarantine directories that are
548 # more than 12 hours old. These are new with Git >= 2.11.0.
549 find -L objects -maxdepth 1 -type d -name 'incoming-?*' -mmin +720 \
550 -exec rm -rf '{}' + || :
552 # Remove any stale shallow_* files that are more than 12 hours old.
553 # These can be left behind by Git >= 1.8.4.2 and < 2.0.0 when a client
554 # requests a shallow clone. Also discard stale .refs-temp* and
555 # .refs-new* files at the same time.
556 find -L . -maxdepth 1 -type f -mmin +720 \( \
557 -name "shallow_?*" -o -name ".refs-temp*" -o -name ".refs-new*" \
558 \) -exec rm -f '{}' + || :
560 # Remove any stale cmbnpcks-* dirs that are more than 12 hours old.
561 # These can be left behind by abnormal exits (e.g. power failure).
562 find -L . -maxdepth 1 -type d -mmin +720 -name "cmbnpcks-?*" \
563 -exec rm -rf '{}' + || :
565 # Remove any stale *.temp files in the objects area that are more than 12 hours old.
566 # This can be stale sha1.temp, or stale *.pack.temp so we kill all stale *.temp.
567 find -L objects -type f -mmin +720 -name "*.temp" -exec rm -f '{}' + || :
569 # Remove any stale *.lock files in the htmlcache area that might have been left
570 # behind after an abnormal exit during an attempt to update a cached file and
571 # are more than 1 hour old.
572 ! [ -d htmlcache ] || find -L htmlcache -type f -mmin +60 -name "*.lock" -exec rm -f '{}' + || :
574 # Remove any stale git-svn temp files that are more than 12 hours old.
575 # The git-svn process creates temp files with random 10 character names
576 # in the root of $GIT_DIR. Unfortunately they do not have a recognizable
577 # prefix, so we just have to kill any files with a 10-character name. We
578 # do this only for git-svn mirrors. All characters are chosen from
579 # [A-Za-z0-9_] so we can at least check that and fortunately the only
580 # collision is 'FETCH_HEAD' but that shouldn't matter.
581 # There may also be temp files with a Git_ prefix as well.
582 if [ -n "$svn_mirror" ]; then
583 _randchar='[A-Za-z0-9_]'
584 _randchar2="$_randchar$_randchar"
585 _randchar4="$_randchar2$_randchar2"
586 _randchar10="$_randchar4$_randchar4$_randchar2"
587 find -L . -maxdepth 1 -type f -mmin +720 -name "$_randchar10" -exec rm -f '{}' + || :
588 find -L . -maxdepth 1 -type f -mmin +720 -name "Git_*" -exec rm -f '{}' + || :
591 # Remove any stale fast_import_crash_<pid> files that are more than 3 days old.
592 if [ -n "$gfi_mirror" ]; then
593 find -L . -maxdepth 1 -type f -mmin +4320 -name "fast_import_crash_?*" -exec rm -f '{}' + || :
596 # Remove any stale core or *.core or core.* files that are more than 3 days old.
597 find -L . -maxdepth 1 -type f -mmin +4320 \( -name "core" -o -name "*.core" -o -name "core.*" \) \
598 -exec rm -f '{}' + || :
602 ## Garbage Collection Types
604 ## There are two kinds of possible garbage collection (gc) operations:
606 ## 1. A normal, full gc
607 ## 2. A "mini" gc
609 ## If the full garbage collection interval has expired (or gc has never been
610 ## run), then a normal, full gc will take place. Otherwise, a "mini" gc will
611 ## take place if the file .needsgc exists.
613 ## A "mini" gc is similar to "git gc --auto" in that it may not end up actually
614 ## doing anything unless the right conditions are present so it's not a burden
615 ## to run it often. If the file .needsgc exists, a "mini" gc will occur at
616 ## the next opportunity.
618 ## See the docs/technical/gc.txt and docs/technical/gc-mini.txt files for more
619 ## of the gory details of how garbage collection is performed.
621 ## Note, however, that the .nogc file suppresses ALL gc activity (normal or mini).
624 proj="${1%.git}"
625 shift
626 cd "$cfg_reporoot/$proj.git"
627 [ -d objects/pack ] || { rm -f gfi-packs; mkdir -p objects/pack; }
628 mirror_url="$(get_mirror_url)" || :
629 svn_mirror=
630 ! is_svn_mirror_url "$mirror_url" || svn_mirror=1
631 gfi_mirror=
632 if [ -f gfi-packs ] && [ -s gfi-packs ] && is_gfi_mirror_url "$mirror_url"; then
633 gfi_mirror=1
636 # If git config --bool --get girocco.redelta is explicitly false then automatic
637 # redelta when there are less than $var_redelta_threshold objects will be suppressed.
638 # On the other hand, if git config --get girocco.redelta is "always" then, on a full
639 # gc only, for the final repack, deltas will always be recomputed.
640 # This can be set on a per-project basis to avoid unusual pathological gc behavior.
641 # Setting this will hurt efficiency of the affected repository.
642 # Note that fast-import packs ALWAYS get new deltas regardless of this setting.
643 noreusedeltaopt="--no-reuse-delta"
644 [ "$(git config --bool --get girocco.redelta 2>/dev/null || :)" != "false" ] || noreusedeltaopt=
645 alwaysredelta=
646 [ "$(git config --get girocco.redelta 2>/dev/null || :)" != "always" ] || alwaysredelta=1
648 # Extract any -f or -F or --no-reuse-object or --no-reuse-delta options
649 # to be compatible with the old and new gc.sh versions and avoid ugly argument
650 # duplication in process lists at the same time
651 # Any options found will override the "girocco.redelta" setting
652 recompress=
653 idx=$#
654 while [ $idx -gt 0 ]; do
655 idx=$(( $idx - 1 ))
656 opt="$1"
657 shift
658 case "$opt" in
659 -f|--no-reuse-delta)
660 alwaysredelta=1
661 continue
663 -F|--no-reuse-object)
664 alwaysredelta=1
665 recompress=1
666 continue
668 -?*)
671 printf >&2 '%s\n' "bad non-option argument: $opt"
672 echo >&2 "(Did you perhaps intend to use a --xxx=yyy form?)"
673 exit 1
674 esac
675 [ -z "$opt" ] || set -- "$@" "$opt"
676 done
677 if [ -n "$alwaysredelta" ]; then
678 noreusedeltaopt="--no-reuse-delta"
679 [ -z "$recompress" ] || noreusedeltaopt="--no-reuse-object"
682 trap 'e=$?; rm -f .gc_in_progress; if [ $e != 0 ]; then echo "gc failed dir: $PWD" >&2; fi' EXIT
683 trap 'exit 130' INT
684 trap 'exit 143' TERM
686 # date -R is linux-only, POSIX equivalent is '+%a, %d %b %Y %T %z'
687 datefmt='+%a, %d %b %Y %T %z'
689 isminigc=
690 if [ "${force_gc:-0}" = "0" ] && check_interval lastgc $cfg_min_gc_interval; then
691 if [ -e .needsgc ]; then
692 isminigc=1
693 else
694 progress "= [$proj] garbage check skip (last at $(config_get lastgc))"
695 exit 0
698 if [ -e .nogc ]; then
699 progress "x [$proj] garbage check disabled"
700 exit 0
702 if ! [ -e .nofetch ] && [ -e .clone_in_progress ] && ! [ -e .clone_failed ]; then
703 progress "x [$proj] garbage check disabled (clone in progress)"
704 exit 0
706 if [ -z "$isminigc" ] && [ -e .delaygc ] && [ -e .needsgc ]; then
707 # Eligible for a full gc but .delaygc is set so it would be skipped
708 # However .needsgc is also set so transform it into a mini instead
709 isminigc=1
710 progress "~ [$proj] garbage check delayed but checking mini because .needsgc"
713 if [ -n "$isminigc" ]; then
714 # Perform a "mini" gc
715 # Note that .delaygc is ignored here as that's only intended for full gc
716 lock_gc
717 rm -f .allowgc .needsgc
718 rm -f objects/pack/pack-*_[rful].keep
719 remove_crud
720 coalesce_reflogs
721 prune_reflogs
722 compact_reflogs
723 maintain_auto_gc_hack
724 generate_auto_gc_update
725 miniactive=
726 if [ -f .needspack ]; then
727 miniactive=1
728 progress "+ [$proj] mini garbage check ($(date))"
729 make_needs_pack
731 if [ -z "$cfg_delay_gfi_redelta" ] && [ -n "$gfi_mirror" ]; then
732 # $Girocco::Config::delay_gfi_redelta is false, force redeltification now
733 if [ -z "$miniactive" ]; then
734 miniactive=1
735 progress "+ [$proj] mini garbage check ($(date))"
737 repack_gfi_packs
739 if lotsa_loose_objects; then
740 if [ -z "$miniactive" ]; then
741 miniactive=1
742 progress "+ [$proj] mini garbage check ($(date))"
744 pack_loose_objects
746 # If there aren't at least 10 non-keep, non-bitmap, non-bndl packs then
747 # don't actually process them yet
748 lpo="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl --quiet"
749 packcnt="$(list_packs --count $lpo objects/pack)" || :
750 if [ "${packcnt:-0}" -ge 10 ]; then
751 if [ -z "$miniactive" ]; then
752 miniactive=1
753 progress "+ [$proj] mini garbage check ($(date))"
755 # if we have at least 10 packs go ahead and pack all refs now too
756 git pack-refs --all --prune
757 if [ -n "$gfi_mirror" ]; then
758 repack_gfi_packs
759 packcnt="$(list_packs --count $lpo objects/pack)" || :
761 # if repack_gfi_packs dropped the pack count to < 10 don't combine
762 if [ "${packcnt:-0}" -ge 10 ]; then
763 combine_small_packs
764 combine_small_loose_packs
765 packcnt="$(list_packs --count $lpo objects/pack)" || :
767 # if we still have more than 10 packs trigger a full gc
768 if [ "${packcnt:-0}" -ge 10 ]; then
769 # We shouldn't be in a .delaygc state at this point, but if
770 # we are then nuke it because we really need a full gc now
771 rm -f .delaygc
772 git config --unset gitweb.lastgc
773 rm -f "$lockf"
774 git update-server-info # just in case
775 progress "- [$proj] mini garbage check triggering full gc too many packs ($(date))"
776 exit 0
779 rm -f "$lockf"
780 if [ -n "$miniactive" ]; then
781 git update-server-info
782 progress "- [$proj] mini garbage check ($(date))"
783 else
784 progress "= [$proj] mini garbage check nothing but crud removal to do ($(date))"
786 exit 0
789 # Avoid unnecessary garbage collections:
790 # 1. If lastreceive is set and is older than lastgc
791 # -AND-
792 # 2. We are not a fork (is_empty_alternates_file) -OR- lastparentgc is older than lastgc
794 # If lastgc is NOT set or lastreceive is NOT set we MUST run gc
795 # If we are a fork and lastparentgc is NOT set we MUST run gc
797 # If the repo is dirty after removing any crud we MUST run gc
799 gcstart="$(date "$datefmt")"
800 skipgc=
801 isfork=
802 is_empty_alternates_file objects/info/alternates || isfork=1
803 lastparentgcsecs=
804 [ -z "$isfork" ] || lastparentgcsecs="$(config_get_date_seconds lastparentgc)" || :
805 lastreceivesecs=
806 if lastreceivesecs="$(config_get_date_seconds lastreceive)" &&
807 [ "${force_gc:-0}" = "0" ] &&
808 lastgcsecs="$(config_get_date_seconds lastgc)" &&
809 [ $lastreceivesecs -lt $lastgcsecs ]; then
810 # We've run gc since we last received, so maybe we can skip,
811 # check if not fork or fork and lastparentgc < lastgc
812 if [ -n "$isfork" ]; then
813 if [ -n "$lastparentgcsecs" ] &&
814 [ $lastparentgcsecs -lt $lastgcsecs ]; then
815 # We've run gc since our parent ran gc so we can skip
816 skipgc=1
818 else
819 # We don't have any alternates (we're not a forK) so we can skip
820 skipgc=1
824 # Prevent any other simultaneous gc operations
825 lock_gc
827 # At this point, if .allowgc or .gc_failed exists, it's now crud to be removed
828 rm -f .allowgc .gc_failed
830 # Always get rid of crud
831 remove_crud
833 # Ideally we would do this in post-receive, but that would mean duplicating the
834 # logic so it's available in the chroot jail and that's highly undesirable
835 # Instead, since the first gc will be triggered immediately following the first
836 # push, we do the check here as it's quick and harmless if HEAD is already valid
837 check_and_set_head || :
839 # Always perform reflogs maintenance
840 coalesce_reflogs
841 prune_reflogs
842 compact_reflogs
844 # Always maintain auto gc hack
845 maintain_auto_gc_hack
846 generate_auto_gc_update
848 # Run 'git svn gc' now for svn mirrors
849 if [ -n "$svn_mirror" ]; then
850 git svn gc || :
853 # Skip the actual gc if .delaygc is set
854 if [ -e .delaygc ]; then
855 progress "x [$proj] garbage check delayed (except for crud removal)"
856 rm -f "$lockf"
857 exit 0
860 # Do not skip gc if the repo is dirty
861 if [ -n "$skipgc" ] && ! is_dirty; then
862 progress "= [$proj] garbage check nothing but crud removal to do ($(date))"
863 config_set lastgc "$gcstart"
864 rm -f "$lockf"
865 exit 0
868 bumptime=
869 if [ -n "$isfork" ] && [ -z "$lastparentgcsecs" ]; then
870 # set lastparentgc and then update gcstart to be at least 1 second later
871 config_set lastparentgc "$gcstart"
872 bumptime=1
874 if [ -z "$lastreceivesecs" ]; then
875 # set lastreceive and then update gcstart to be at least 1 second later
876 config_set lastreceive "$gcstart"
877 bumptime=1
879 if [ -n "$bumptime" ]; then
880 sleep 1
881 gcstart="$(date "$datefmt")"
884 progress "+ [$proj] garbage check ($(date))"
886 newdeltas=
887 [ -z "$alwaysredelta" ] || newdeltas="$noreusedeltaopt"
888 if [ -z "$newdeltas" ] && [ -n "$gfi_mirror" ]; then
889 if [ $(list_packs --exclude-no-idx --count objects/pack) -le \
890 $(list_packs --exclude-no-idx --count --quiet --only gfi-packs) ]; then
891 # Don't bother with repack_gfi_packs since everything's being repacked
892 newdeltas="--no-reuse-delta"
895 if [ -z "$newdeltas" ] && [ -n "$noreusedeltaopt" ] &&
896 [ $(list_packs --all --exclude-no-idx --count-objects objects/pack) -le $var_redelta_threshold ]; then
897 # There aren't enough objects to worry about so just redelta to get the best pack
898 newdeltas="--no-reuse-delta"
900 if [ -z "$newdeltas" ]; then
901 # Since we're not going to recompute deltas overall, we need to do the
902 # "mini" maintenance so that we can get more optimal deltas
903 [ -z "$noreusedeltaopt" ] || make_needs_pack
904 repack_gfi_packs
905 force_single_pack_redelta=
906 [ -n "$gfi_mirror" ] || [ -n "$svn_mirror" ] || force_single_pack_redelta=1
907 [ -z "$noreusedeltaopt" ] || combine_small_packs $force_single_pack_redelta
908 [ -z "$noreusedeltaopt" ] || combine_small_loose_packs $force_single_pack_redelta
912 ## Safe Pruning In Forks
914 ## We are about to perform garbage collection. We do NOT use the "git gc" or
915 ## the "git repack" commands directly as they do not provide enough control over
916 ## the fine details. However, we DO maintain a "gc.pid" file during our garbage
917 ## collection so that a simultaneous "git gc" by an administrator will be
918 ## blocked (and similarly we refuse to start garbage collection if we cannot
919 ## create the "gc.pid" file).
921 ## When we say "gc" in the below description we are referring to our "gc.sh"
922 ## script, NOT the "git gc" command.
924 ## If the project we are running garbage collection (gc) on has any forks we
925 ## must be careful not to remove any objects that while no longer referenced by
926 ## this project (the parent) are still referenced by one or more forks (the
927 ## children) otherwise the children will become corrupt and we can't abide
928 ## corrupt children.
930 ## One way to accomplish this is to simply hard-link all currently existing
931 ## loose objects and packs in the parent into all the children that refer to the
932 ## parent (via a line in their objects/info/alternates file) before beginning
933 ## the gc operation and then relying on a subsequent gc in the child to clean up
934 ## any excess objects/packs. We used to use this strategy but it's very
935 ## inefficient because:
937 ## 1. The disk space used by the old pack(s)/object(s) will not be reclaimed
938 ## until all children (and their children, if any) run gc by which time
939 ## it's quite possible the topmost parent will have run gc again and
940 ## hard-linked yet another old pack down to its children (not to mention
941 ## loose objects).
943 ## 2. When using the "-A" option with "git repack", any new objects in the
944 ## parent that are not referenced by children will continually get
945 ## exploded out of the hard-linked pack in the children whenever the
946 ## children run gc.
948 ## 3. To avoid suboptimal and/or unnecessarily many packs being hard-linked
949 ## into child forks, we must run the "mini" gc maintenance before we
950 ## perform the hard-linking into the children which provides yet another
951 ## source of inefficiency.
953 ## While we were still using the "-A" option to "git repack" (that was not
954 ## always the case) to guarantee we can access old ref values for long enough
955 ## to send out a meaningful mail.sh notification, another, more efficient,
956 ## option became available to prevent corruption of child forks that continue
957 ## to refer to objects that are no longer reachable from any ref in the parent.
959 ## The only things that need be copied (or hard-linked) into the child fork(s)
960 ## are those objects that have become unreachable from any ref in the parent.
962 ## When we were using the "git repack -A -d" + "git prune --expire=1.day.ago"
963 ## technique, the only objects that could ever be removed were loose objects
964 ## that "git prune" determined were expired. In that case, loose objects were
965 ## all that need be hard-linked down to child forks in order to avoid
966 ## corruption of any child fork(s).
968 ## The "git repack -A -d" + "git prune --expire=1.day.ago" + hard-linking loose
969 ## objects to child forks technique remains fundamentally sound from the
970 ## perspective of supporting simultaneous gc and push and keeping newly
971 ## unreachable objects around long enough to be sure we can send out meaningful
972 ## ref change notifications and never corrupting any child forks and never
973 ## persisting the lifetime of large old packs containing mostly duplicate or
974 ## unreachable objects as gc percolates through a project's entire fork tree.
976 ## However, that technique suffers from one potential prodigious pitfall.
978 ## Unreachable objects come flying out of their packs to splatter all over the
979 ## objects subdirectories possibly creating a huge, inefficient mess.
981 ## Often this is not an issue. Even with a lot of rebasing going on, usually
982 ## the only objects that will splatter are some commits, trees and the odd blob
983 ## here and there. Not enough to be overly concerned about.
985 ## However, for the reppository that frequently experiences a lot of non-fast-
986 ## forward updates and/or outright ref deletion, the number of objects suddenly
987 ## popping out of their packs at "git repack -A -d" time can be overwhelming.
989 ## To avoid this issue we now use a four phase pack creation strategy.
990 ## This will result in creation of up to four packs (instead of at most one).
992 ## I. A complete pack (with bitmaps if appropriate) gets created including
993 ## only "reachable" objects from all refs/... refs plus HEAD. This will
994 ## also serve as the virtual bundle for the repository.
996 ## II. A pack of recently-became-unreachable objects and friends is created.
997 ## (The "friends" are ref logs, linked working tree HEADs and indicies.)
998 ## Because both the pre-receive and update.sh script record all ref
999 ## changes we can easily choose the cut off point for "recently".
1000 ## It is only the fact we maintain those logs in the reflogs subdirectory
1001 ## that allows this step to be possible.
1003 ## III. If the repository has any forks with a non-zero length alternates file,
1004 ## yet another pack of "--keep-unreachable" objects is generated that will
1005 ## not actually be kept in the parent, but hard-linked into all the forks.
1007 ## IV. Finally, after running "git prune-packed", any remaining loose objects
1008 ## are migrated into a pack of their own.
1010 ## We then remove any non-.keep packs that existed before we started the
1011 ## process being careful to keep any same-pack pushes for the "Push Pack Redux"
1012 ## race condition (see docs/technical/gc.txt).
1014 ## By using "git pack-objects" directly we are able to accomplish this with
1015 ## very little additional effort.
1017 ## The packs produced by (III) are treated almost like ".keep" packs by child
1018 ## forks in that the objects in them are never repacked into any other
1019 ## "--keep-unreachable" packs (but they can migrate into phase I or II packs)
1020 ## and those phase III packs are then hard-linked into any grandchild forks.
1022 ## This avoids the space explosion that could occur if each fork level ended
1023 ## up duplicating the "--keep-unreachable" pack space by repacking those
1024 ## objects (essentially breaking the hard-link to the single copy of those
1025 ## objects).
1027 ## While it is true that each level of forks could potentially add yet another
1028 ## phase III pack to be hard-linked down to its children, such packs will only
1029 ## include unreachable objects not already in any phase III packs that were
1030 ## received from the parent.
1032 ## The space for the phase III packs will not be reclaimed until the gc
1033 ## finishes percolating through the entire "fork tree" of a project.
1035 ## This is not much different than the "git repack -A -d" situation where
1036 ## all the loose objects are hard-linked down into child forks. In that
1037 ## case forks that actually need any of those objects could gradually reduce
1038 ## the number of objects hard-linked into deeper fork levels.
1040 ## The difference with a phase III "--keep-unreachable" pack is that there
1041 ## cannot be any gradual reduction like that since it would require repacking
1042 ## the pack and breaking the hard-link thereby increasing storage space. The
1043 ## storage will instead always be reclaimed all at once when all of the
1044 ## projects in the "fork tree" complete their gc.
1046 ## However, the belief is that the huge space win by having all the
1047 ## unreachable objects packed up together far eclipses (when many objects are
1048 ## involved, the single-pack version can end up using 1/20th or less of the
1049 ## disk space compared to having them all as loose objects) any brief minor
1050 ## space savings that might occur under the "git repack -A -d" loose object
1051 ## system prior to the gc collection completing for all the projects in the
1052 ## "fork tree".
1056 ## utility functions
1059 make_packs_ugw() {
1060 find -L "$1" -maxdepth 1 -type f ! -perm -ug+w \
1061 -name "pack-$octet20*.pack" -exec chmod ug+w '{}' + || :
1062 } 2>/dev/null
1064 get_index_tree() {
1065 if [ -s "$1" ]; then
1066 GIT_INDEX_FILE="$1"
1067 export GIT_INDEX_FILE
1068 git write-tree 2>/dev/null || :
1069 unset GIT_INDEX_FILE
1073 get_detached_head() {
1074 if [ -s "$1" ] && read -r _head <"$1" 2>/dev/null; then
1075 case "$_head" in $octet20*)
1076 echo "$_head"
1077 esac
1081 # single argument must be a top-level GIT_DIR
1082 # output will be (one per line, zero or more lines) full hash
1083 # values (i.e. 40 or more hex digits) from possible detached head
1084 # "single level" refs (e.g. "FETCH_HEAD" "MERGE_HEAD" etc.) from
1085 # files in the specified directory that have a "suitable" name and
1086 # have a length >= 40 and <= 1000 where EVERY line in a file MUST
1087 # match a 40 digit (or longer) hex string or that file will be ignored.
1088 # In a nod to Git each line first has any tab and anything following truncated.
1089 # NO sorting NOR "uniq"ing NOR "--batch-check"ing is performed on the output.
1090 # Also note that "HEAD" is NOT EXCLUDED and if detached will be output!
1091 get_detached_friends() {
1092 if [ -n "$1" ] && [ -d "$1" ]; then
1093 LC_ALL=C find -H "$1" -maxdepth 1 -name '[A-Za-z]*[A-Za-z0-9]' \
1094 -type f -size +39c -size -1001c -exec awk '
1095 BEGIN {exit}
1096 function process(fn, fnt, hashes, hcnt, fl, hi) {
1097 fnt = fn; sub(/^.*\//, "", fnt)
1098 if (length(fnt) > 31 || fnt !~ /^[A-Za-z][A-Za-z0-9_-]*[A-Za-z0-9]$/) return;
1099 hcnt = 0;
1100 while (getline fl < fn) {
1101 sub(/\t.*$/, "", fl)
1102 if (length(fl) < 40 || fl !~ /^[0-9a-fA-F][0-9a-fA-F]*$/) {hcnt = 0; break;}
1103 hashes[++hcnt] = fl;
1105 close(fn)
1106 for (hi=1;hi<=hcnt;++hi) print tolower(hashes[hi]);
1108 END {for (idx=1;idx<ARGC;++idx) process(ARGV[idx])}
1109 ' '{}' '+' 2>/dev/null || :
1113 # get_worktrees_friends
1114 # single argument is a (possibly relative) path to a "worktrees" subdir
1115 # if omitted it defaults to "worktrees"
1116 # any "friends" found in there are output to stdout
1117 get_worktrees_friends() {
1118 if [ -d "${1:-worktrees}" ]; then
1119 find -L "${1:-worktrees}" -mindepth 2 -maxdepth 2 -name HEAD -type f -print |
1120 while read -r _lwth; do
1121 get_detached_head "$_lwth"
1122 get_detached_friends "${_lwth%HEAD}"
1123 get_index_tree "${_lwth%HEAD}index"
1124 done
1128 # compute_extra_reachables
1129 # create lines suitable for a packed-refs file mentioning all the
1130 # other refs we might like to keep.
1131 # the current directory MUST be set to the repository's --git-dir
1132 # the following are included:
1133 # * refs mentioned in repack/packed-refs.extra (if it exists)
1134 # * refs mentioned in reflogs/... files
1135 # * tree(s) created from index file(s)
1136 # * detached linked working tree heads
1137 # Resulting objects are tested for existence and uniqified then output
1138 # one per line under a refs/z* namespace
1139 compute_extra_reachables() {
1141 if [ -s repack/packed-refs.extra ]; then
1142 LC_ALL=C sed <repack/packed-refs.extra -n \
1143 -e 's/^\([0-9A-Fa-f][0-9A-Fa-f]*\).*$/\1/p' \
1144 -e 's/^\^\([0-9A-Fa-f][0-9A-Fa-f]*\).*$/\1/p'
1146 digits8='[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]'
1147 find -L reflogs -mindepth 1 -maxdepth 1 -type f -name "$digits8*" -exec gzip -c -d -f '{}' + |
1148 LC_ALL=C awk '{print $2; print $3}'
1149 get_detached_friends .
1150 ! [ -f index ] || get_index_tree index
1151 get_worktrees_friends
1153 is_git_dir private &&
1154 [ "$(cd objects && pwd -P)" = "$(cd private/objects && pwd -P)" ] &&
1155 [ "$(cd refs && pwd -P)" != "$(cd private/refs && pwd -P)" ]
1156 then
1157 git --git-dir=private show-ref --head --hash 2>/dev/null || :
1158 get_detached_friends private
1159 ! [ -f private/index ] || get_index_tree private/index
1160 get_worktrees_friends private/worktrees
1162 } | LC_ALL=C sort -u |
1163 git cat-file ${var_have_git_260:+--buffer} --batch-check"${var_have_git_185:+=%(objectname)}" |
1164 LC_ALL=C awk '!/missing/ {num++; print $1 " " "refs/" substr("zzzzzzzzzzzz", 1, length(num)) "/" num}'
1168 ## main gc logic
1171 # Everything else is more efficient if we do this first
1172 # The "--prune" option is the default since v1.5.0 but it serves as "documentation" here
1173 git pack-refs --all --prune
1174 [ -e packed-refs ] || >>packed-refs # should never happen...
1176 # We didn't used to do anything about rerere or worktrees but we're
1177 # trying to make nice with linked working trees these days :)
1178 # Maybe even non-bare repositories too, but *shush* about those ;)
1180 # If we have a worktrees subdir and are running Git >= 2.5.0, prune worktrees
1181 # By doing this first we may save some work if any worktrees are pruned
1182 # Git will remove the worktrees subdir entirely when the last worktree is pruned
1183 if [ -n "$var_have_git_250" ] && [ -d worktrees ]; then
1184 # The value "3.months.ago" is hard-coded into gc.c rather than
1185 # having the default be in worktree.c so we must provide it if
1186 # we get nothing out of the gc.worktreePruneExpire config item
1187 # Prior to Git v2.6.0 the config item was gc.pruneworktreesexpire
1188 # however we just always use the newer name no matter what Git version
1189 expiry="$(git config --get gc.worktreePruneExpire 2>/dev/null)" || :
1190 eval git worktree prune --expire '"${expiry:-3.months.ago}"' "${quiet:+>/dev/null 2>&1}" || :
1193 # If we have a logs directory or a worktrees directory expire the ref logs now
1194 # Note that Git itself does not use either --rewrite or --updateref, so neither do we
1195 ! [ -d logs ] && ! [ -d worktrees ] || eval git reflog expire --all "${quiet:+>/dev/null 2>&1}" || :
1197 # git rerere does it right and handles its own default/config'd expiration values
1198 ! [ -d rr-cache ] || eval git rerere gc "${quiet:+>/dev/null 2>&1}" || :
1200 make_repack_dir
1201 ! [ -e .gc_failed ] || exit 1
1202 rm -f .gc_in_progress # make sure
1203 touch .gc_in_progress # it's truly fresh
1204 rm -f bundles/* objects/pack/pack-*.bndl
1205 # These only exist for a brief time before the packs loose their _f suffix
1206 # "Push Pack Redux" does not apply to these since they were only ever present with _f
1207 rm -f objects/pack/pack-*_f.keep
1208 # This is perhaps a bit aggressive in that if we're suffering from "Push Pack Redux"
1209 # and somehow we get run again immediately after the run where "Push Pack Redux" happened
1210 # and we have garbage collection forced, there's just the barest, almost negligible,
1211 # possibility that the "Push Pack Redux" ref updates _still_ have not happened and we
1212 # should not be removing _r .keep files. None of the normal Girocco processing can
1213 # cause this. The second run of this script would have to use the force gc option
1214 # for it to even be possible in the first place. What's much more likely is that
1215 # the initial run of this script was somehow interrupted in the middle before it
1216 # could get rid of the _r .keep file itself in which case it's better to get rid of
1217 # it now to avoid keeping something around that would perturb our nice and neat gc
1218 rm -f objects/pack/pack-*_r.keep
1219 # We will add .keep files for _u and _l packs if and when we run phase III
1220 # Otherwise they need to not have any .keep files during phases I and II
1221 rm -f objects/pack/pack-*_[ul].keep
1223 # We need to make sure that any non-Girocco (barely tolerated) Git object creation
1224 # activity will be able to "freshen" the pack containing a pre-existing object
1225 # that's being written. This really should not be necessary as the pre-receive
1226 # hook should make sure this takes place for any incoming pushes.
1227 # However, do it here anyway just in case.
1228 make_packs_ugw objects/pack
1230 # This is only effective with Git v2.3.5 and later and it will only matter when
1231 # we are using one of the "internal_rev_list" modes of pack-objects
1232 # (the combine-packs.sh script never uses any of those modes)
1233 # The "git repack" and "git prune" commands always set this internally themselves
1234 # It makes no difference if there's no repository corruption
1235 GIT_REF_PARANOIA=1 && export GIT_REF_PARANOIA
1237 # All of the options we might want to use with pack-objects were supported
1238 # at some point prior to Git version v1.6.6 which is the minimum version that
1239 # Girocco now requires. Except for one (--use-bitmap-index). Several of them
1240 # are "boiler plate" options we always want to use so we bundle them up here.
1241 pkopt="--delta-base-offset --keep-true-parents --non-empty --all-progress-implied"
1242 # We want to use --include-tag, but before Git v2.10.1 it would leave out
1243 # "middle" tags (e.g. a tag of a tag of a commit would omit the tagged tag)
1244 # See http://repo.or.cz/git.git/b773ddea2cd3b08c for details
1245 # ("pack-objects: walk tag chains for --include-tag", 2016-09-07, v2.10.1)
1246 # This is not a free check as it matches all refs against refs/tags/ then
1247 # peels all the annotated tags and checks for inclusion. The situation in
1248 # which it would add a tag that was not already included by a reachability
1249 # trace that included tag starting points can only occur if a new tag gets
1250 # pushed during gc pointing to something that would have been packed anyway.
1251 # But, it could happen and, really, compared to gc as a whole it's not that
1252 # expensive to perform (provided we do not get an unconnected pack).
1253 [ -z "$var_have_git_2101" ] || pkopt="$pkopt --include-tag"
1254 pkopt="$pkopt ${quiet:---progress} $packopts"
1256 # The git pack-objects command only supports bitmaps if all objects are being
1257 # packed (the "--all" option) and the "--stdout" option is NOT being used.
1258 # Additionally, while packing, if any encountered reachable objects are
1259 # determined to be "not wanted" then no bitmap index will be written anyway.
1260 # While it is theoretically possible that a project with a non-empty alternates
1261 # file ends up packing all objects (because it does not actually use any of the
1262 # objects found in the alternates), it's very unlikely. And, in the unlikely
1263 # event that did occur, clients would see a message about only using one bitmap
1264 # because Git can only use one bitmap at a time and at least one of the
1265 # alternates is bound to have a bitmap. Therefore if we see a non-empty
1266 # alternates file, we disable writing bitmaps which avoids the warning and any
1267 # possibility of a client warning as well. Also if we are running anything
1268 # before Git v2.1.0 (the effective version for repack.writeBitmaps=true) then
1269 # we also always disable bitmap writing.
1270 wbmopt=
1271 [ -z "$var_have_git_210" ] || wbmopt="--write-bitmap-index"
1272 # More recent versions of pack-objects have optimizations when not using the
1273 # --local option. If we do not have any alternates it's a pointless option.
1274 # If we do have alternates we need to skip writing a bitmap and we cannot
1275 # have a bundle since it must contain all objects.
1276 if [ -n "$isfork" ]; then
1277 lclopt="--local"
1278 wbmopt=
1279 makebndl=
1280 else
1281 lclopt=
1282 makebndl=1
1286 ## Phase I
1289 wbmstr=
1290 [ -n "$wbmopt" ] || wbmstr=" (bitmaps disabled)"
1291 progress "~ [$proj] running primary full gc pack-objects$wbmstr ($(date))"
1293 gotforks=
1294 ! has_forks_with_alternates "$proj" || gotforks=1
1296 # To avoid "Push Pack Redux" (see docs/technical/gc.txt), after collecting the
1297 # initial preexisting non-keep pack list, we rename them so that an incoming push
1298 # pack cannot possibly experience a pack name collision. Git does not require
1299 # use of the "default" pack names, simply that the proper extensions are used.
1300 # We rename to insert an "_r" just before the extension to avoid "Push Pack Redux"
1301 # name collisions. Later on we may create an "unreachable" pack for hard-linking
1302 # down into forks and it will have an "_u" inserted just before its extension.
1303 packlist="$(list_packs -C objects/pack --all --exclude-no-idx --exclude-keep --quiet .)" || :
1304 oldpacks=
1305 for oldpack in $packlist; do
1306 oldpack="${oldpack%.pack}"
1307 [ -f "objects/pack/$oldpack.pack" ] || {
1308 echo >&2 "[$proj] unable to list old pack files"
1309 exit 1
1311 case "$oldpack" in pre-auto-gc-[12])
1312 # we never disturb pre-auto-gc-1 or pre-auto-gc-2 packs
1313 continue
1314 esac
1315 oldpackhex="${oldpack#pack-}"
1316 if [ "${oldpackhex#*[!0-9a-fA-F]}" != "$oldpackhex" ]; then
1317 # names not exclusively hexadecimal do not need renaming
1318 case "$oldpack" in
1319 pack-$octet20*_l)
1320 # _l packs are treated like still-unpacked loose objects
1321 continue;;
1322 *_f)
1323 # _f packs can only be left over from a previously interrupted gc;
1324 # they need to be renamed to _r now so they're not confused with
1325 # any freshly generated "final" packs (and we already removed
1326 # any pre-existing *_f.keep files so we're good to go)
1329 oldpacks="${oldpacks:+$oldpacks }$oldpack"
1330 continue;;
1331 esac
1333 rename_pack "objects/pack/$oldpack" "objects/pack/${oldpack%_f}_r" || {
1334 echo >&2 "[$proj] unable to rename old pack files"
1335 exit 1
1337 # If the oldpack has a .keep now it means a "Push Pack Redux" is actually
1338 # in progress at this moment and we need to .keep the renamed pack,
1339 # otherwise no "Push Pack Redux" has started yet or it has already finished.
1340 # In either case we're okay because if it's just finished then all ref
1341 # changes have already been made so we don't need a .keep and we will
1342 # see the ref changes and grab all the objects via a reachability trace.
1343 # If it hasn't started yet that's okay because we're done moving that
1344 # name so a complete pack will appear under the old name that we'll
1345 # leave alone.
1346 if [ -f "objects/pack/$oldpack.keep" ]; then
1347 echo "Push Pack Redux" >"objects/pack/${oldpack%_f}_r.keep"
1348 else
1349 oldpacks="${oldpacks:+$oldpacks }${oldpack%_f}_r"
1351 done
1353 # We wish to keep deltas from our last full pack so if we're not redeltaing
1354 # then make sure the .pack associated with the .bitmap has a newer mod time
1355 # (If there is no .bitmap then touch the pack with the most objects instead.)
1356 if [ -z "$newdeltas" ]; then
1357 bmpack="$(list_packs --exclude-no-bitmap --exclude-no-idx --max-matches 1 objects/pack)"
1358 [ -n "$bmpack" ] || bmpack="$(list_packs --exclude-no-idx --max-matches 1 --object-limit -1 --include-boundary objects/pack)"
1359 if [ -n "$bmpack" ] && [ -f "$bmpack" ] && [ -s "$bmpack" ]; then
1360 sleep 1
1361 touch -c "$bmpack" 2>/dev/null || :
1362 # We must touch .gc_in_progress here to avoid $bmpack looking
1363 # like it's been "freshened" when redundant packs are removed
1364 # It's okay if they have the same mod time, but POSIX does not
1365 # guarantee an ordering for the "touching" that occurs which is
1366 # why this must be a separate command but needs no "sleep 1"
1367 touch .gc_in_progress
1371 # Now we need to make sure that any "freshening" that takes place will actually
1372 # result in a "newer" modification time than the .gc_in_progress file now has
1373 sleep 1
1375 # We run git pack-objects from the repack subdirectory so we can force
1376 # optimized packs to be generated even for repositories that do not have any
1377 # tagged commits
1378 packs="$(git --git-dir=repack pack-objects </dev/null \
1379 $pkopt --all $newdeltas $lclopt ${wbmopt:---honor-pack-keep} "$@" repack/alt/pack/pack)"
1380 v_cnt packcnt $packs
1381 [ $packcnt -eq 1 ] || makebndl=
1384 ## Phase II
1387 progress "~ [$proj] running supplementary gc pack-objects ($(date))"
1389 # Add the "supplementary" refs
1390 compute_extra_reachables >>repack/packed-refs
1392 # Subtract the primary refs
1393 GIT_ALTERNATE_OBJECT_DIRECTORIES="$PWD/repack/alt"
1394 export GIT_ALTERNATE_OBJECT_DIRECTORIES
1396 # For this one we MUST use --local and MUST NOT use --write-bitmap-index
1397 # However, if there is a "logs" subdirectory we need to use --reflog
1398 # We do add it, just in case, if the linked working trees dir is present
1399 # We do not add --indexed-objects as that requires v2.2.0 and it's unclear
1400 # if it properly includes linked working tree index files or not. The
1401 # above compute_extra_reachables has already included all index trees (thereby
1402 # providing proper --indexed-objects support for all Git versions) making the
1403 # option completely unnecessary.
1404 rflopt=
1405 ! [ -d logs ] && ! [ -d worktrees ] || rflopt=--reflog
1406 spacks="$(git --git-dir=repack pack-objects </dev/null \
1407 $pkopt --honor-pack-keep --all $rflopt $newdeltas --local "$@" repack/alt/pack/pack)"
1410 ## Phase III
1413 # There's nothing to do for Phase III unless we have forks that refer to our
1414 # project from their alternates file
1415 hlpacks=
1416 upacks=
1417 if [ -n "$gotforks" ]; then
1419 progress "~ [$proj] running keep-unreachable gc pack-objects for forks ($(date))"
1421 # If we are a fork, any pre-existing _u packs need to have a .keep
1422 # for this phase and be added to the hlpacks list otherwise (we are
1423 # not a fork) pre-existing _u packs are anomalies to be treated like
1424 # regular non-_u packs
1425 if [ -n "$isfork" ]; then
1426 for upack in $(find -L objects/pack -mindepth 1 -maxdepth 1 -name "pack-$octet20*_[ul].pack" -print); do
1427 upack="${upack%.pack}"
1428 [ -e "$upack.keep" ] || echo "unreachable" >"$upack.keep"
1429 case "$upack" in *_l);;*)
1430 hlpacks="${hlpacks:+$hlpacks }${upack#objects/pack/pack-}"
1431 esac
1432 done
1434 # Using either --no-reuse-delta or --no-reuse-object together with the
1435 # --keep-unreachable option is a very, very, very bad idea when good
1436 # packs are the desired outcome. If newdeltas are being generated
1437 # then we pack to a temp name, and use combine-packs.sh to get a better
1438 # pack as the result to avoid making a bad --keep-unreachable pack
1439 pfx=
1440 [ -z "$newdeltas" ] || pfx="ku"
1441 upacks="$(git --git-dir=repack pack-objects </dev/null \
1442 $pkopt --honor-pack-keep --all $rflopt --keep-unreachable --local "$@" repack/alt/pack/${pfx}pack)"
1443 if [ -n "$upacks" ] && [ -n "$newdeltas" ]; then
1444 progress "~ [$proj] rebuilding keep-unreachable pack deltas"
1445 oldupacks="$upacks"
1446 upacks="$(
1447 printf "repack/alt/pack/${pfx}pack-%s.pack\n" $oldupacks |
1448 run_combine_packs --names --weak-naming --non-empty --all-progress-implied ${quiet:---progress} \
1449 $packopts $newdeltas "$@" repack/alt/pack/pack)"
1450 eval rm -f "$(printf \""repack/alt/pack/${pfx}pack-%s.*"\"" " $oldupacks)"
1452 for upack in $upacks; do
1453 rename_pack "repack/alt/pack/pack-$upack" "repack/alt/pack/pack-${upack}_u"
1454 done
1455 rm -f objects/pack/pack-*_[ul].keep
1456 [ -z "$hlpacks" ] && [ -z "$upacks" ] ||
1457 progress "~ [$proj] hard-linking keep-unreachable pack(s) into immediate child forks"
1459 # We have to update the lastparentgc time in the child forks even if they do not get any
1460 # new "unreachable packs" because they need to run gc just in case the parent now has some
1461 # objects that used to only be in the child so they can be removed from the child.
1462 # For example, a "patch" might be developed first in a fork and then later accepted into
1463 # the parent in which case the objects making up the patch in the child fork are now
1464 # redundant (since they're now in the parent as well) and need to be removed from the
1465 # child fork which can only happen if the child fork runs gc.
1466 lastparentgc="$(date "$datefmt")"
1468 # It is enough to copy objects just one level down and get_repo_list
1469 # takes a regular expression (which is automatically prefixed with '^')
1470 # so we can easily match forks exactly one level down from this project
1471 forkdir="$proj"
1472 get_repo_list "$forkdir/[^/:][^/:]*:" |
1473 while read fork; do
1474 # Ignore forks that do not exist or are symbolic links
1475 ! [ -L "$cfg_reporoot/$fork.git" ] && [ -d "$cfg_reporoot/$fork.git" ] ||
1476 continue
1477 # Or have an empty alternates file
1478 ! is_empty_alternates_file "$cfg_reporoot/$fork.git/objects/info/alternates" ||
1479 continue
1480 runupdate=
1481 # Match hlpacks in parent project if any
1482 if [ -n "$hlpacks" ]; then
1483 mkdir -p "$cfg_reporoot/$fork.git/objects/pack"
1484 eval ln -f "$(printf '"objects/pack/pack-%s.pack" ' $hlpacks)" \
1485 "$(printf '"objects/pack/pack-%s.idx" ' $hlpacks)" \
1486 '"$cfg_reporoot/$fork.git/objects/pack/"'
1487 runupdate=1
1489 # Match upacks in repack/alt area if any
1490 if [ -n "$upacks" ]; then
1491 mkdir -p "$cfg_reporoot/$fork.git/objects/pack"
1492 eval ln -f "$(printf '"repack/alt/pack/pack-%s_u.pack" ' $upacks)" \
1493 "$(printf '"repack/alt/pack/pack-%s_u.idx" ' $upacks)" \
1494 '"$cfg_reporoot/$fork.git/objects/pack/"'
1495 runupdate=1
1497 if ! [ -e "$cfg_reporoot/$fork.git/.needsgc" ]; then
1498 # Trigger a mini gc in the fork if it now has too many packs
1499 fpackcnt="$(list_packs --quiet --count --exclude-no-idx --exclude-keep "$cfg_reporoot/$fork.git/objects/pack")" || :
1500 if [ -n "$fpackcnt" ] && [ "$fpackcnt" -ge 20 ]; then
1501 >"$cfg_reporoot/$fork.git/.needsgc"
1504 [ -z "$runupdate" ] || git --git-dir="$cfg_reporoot/$fork.git" update-server-info
1505 # Update the fork's lastparentgc date (must be more recent than $gcstart)
1506 git --git-dir="$cfg_reporoot/$fork.git" config gitweb.lastparentgc "$lastparentgc"
1507 done
1510 # Now move any primary/supplementary packs back into objects/pack
1511 # then drop any "unfreshened" redundant packs and clear repack/alt
1513 # First make sure the primary pack(s) have the most recent mod time
1514 if [ -n "$packs" ]; then
1515 [ -z "$spacks" ] || sleep 1
1516 printf 'repack/alt/pack/pack-%s.pack\n' $packs | xargs touch -c 2>/dev/null || :
1519 # Move the packs into place but with a _f suffix and a .keep file for now
1520 for pack in $packs $spacks; do
1521 rename_pack "repack/alt/pack/pack-$pack" "objects/pack/pack-${pack}_f"
1522 [ -e "objects/pack/pack-${pack}_f.keep" ] ||
1523 echo "final" >"objects/pack/pack-${pack}_f.keep"
1524 done
1526 # It's possible that one of the $oldpacks had a .bitmap, got renamed (along
1527 # with its .bitmap) and then got "freshened" causing us to not remove it
1528 # However, if $wbmopt is set we most likely now have TWO .bitmap packs!
1529 # This can produce ugly warnings we don't want and possibly get the wrong
1530 # bitmap used since only one .bitmap file can ever be used by Git.
1531 # If this has happened, the .bitmap we want to discard will always have
1532 # an _r suffix so we can just zap any such now since it will leave the pack.
1533 [ -z "$wbmopt" ] || rm -f objects/pack/pack-*_r.bitmap || :
1535 # Remove the redundant packs that have not since been "freshened"
1536 # This does not completely eliminate the race condition window (Girocco's own
1537 # activites -- gc/fetch/receive are immune to the race) but it substantially
1538 # shrinks it down to just the time after the find but before the following rm
1539 >repack/oldpacks
1540 [ -z "$oldpacks" ] ||
1541 printf 'objects/pack/%s.pack\n' $oldpacks |
1542 LC_ALL=C sort >repack/oldpacks
1543 find -L objects/pack -maxdepth 1 -type f -name "pack-$octet20*.pack" -newer .gc_in_progress -print |
1544 LC_ALL=C sort >repack/freshened
1545 deadpacks="$(LC_ALL=C join -v 1 repack/oldpacks repack/freshened | LC_ALL=C sed 's/\.pack$//')"
1546 [ -z "$deadpacks" ] ||
1547 eval echo "$(printf '"%s".* ' $deadpacks)" | xargs rm -f || :
1549 # No need for this anymore
1550 rm -rf repack/alt objects/pack/repack
1551 unset GIT_ALTERNATE_OBJECT_DIRECTORIES
1554 ## Phase IV
1557 progress "~ [$proj] running gc prune-packed"
1559 # We do not want the redundant packs or any new "--keep-unreachable" pack(s) to be
1560 # present while running prune-packed. We try to guarantee that any loose object
1561 # (or any object present in a pack with an _l suffix which was created by mini gc)
1562 # that's unreachable persists for at least one $Girocco::Config::min_gc_interval
1563 # (not withstanding administrator interference to force earlier gc to occur).
1564 # If we were to include the redundant/keep-unreachable pack(s) when running
1565 # prune-packed and a loose unreachable object happened to be duplicated in one
1566 # of them we would end up removing it too soon and void our guarantee.
1567 git prune-packed $quiet
1569 progress "~ [$proj] running loose objects gc pack-objects ($(date))"
1571 # Although Git v2.10.0 and later support a --pack-loose-unreachable option,
1572 # we MUST NOT use it for these reasons:
1573 # 1) We're not interested in expensive "unreachable" at this point, only "loose"
1574 # 2) It produces simply horrid packs about 3.8x times larger than they should be
1575 # 3) We don't require anything more than Git v1.6.6
1576 # The only way we could see any _o pack files at this point is if one got
1577 # "freshened" while we were running gc. If that happens then it gets to live on
1578 # until the next full gc and we need to include it in the loose repack here.
1579 lpacks="$(list_packs --exclude-no-idx --exclude-no-sfx _l --exclude-no-sfx _o --quiet objects/pack |
1580 run_combine_packs --replace --names --loose --weak-naming --non-empty --honor-pack-keep \
1581 --all-progress-implied ${quiet:---progress} $packopts $newdeltas "$@")"
1583 if [ -n "$lpacks" ]; then
1584 # Make sure any primary pack(s) have a more recent mod time than "unreachable" objects packs
1585 if [ -n "$packs" ]; then
1586 sleep 1
1587 printf 'objects/pack/pack-%s_f.pack\n' $packs | xargs touch -c 2>/dev/null || :
1589 # We need to identify these packs later so we don't combine_packs them
1590 for objpack in $lpacks; do
1591 rename_pack "objects/pack/pack-$objpack" "objects/pack/pack-${objpack}_o" || :
1592 done
1595 # Polish up the final packs now
1596 rm -f objects/pack/pack-*_f.keep
1597 for pack in $packs $spacks; do
1598 rename_pack "objects/pack/pack-${pack}_f" "objects/pack/pack-$pack"
1599 done
1601 if [ -n "$lpacks" ]; then
1602 # Finally zap the corresponding loose objects
1603 progress "~ [$proj] running packed loose objects gc prune-packed"
1604 git prune-packed $quiet
1607 ! [ -e .gc_failed ] || exit 1
1608 # These, if they exist, are now meaningless and need to be removed
1609 rm -f gfi-packs .needsgc .needspack .needspackgc
1611 # Make sure this stays up to date
1612 git update-server-info
1614 # We must make loose objects group writable so that they
1615 # can be freshened by other pushers. Technically we need only do this for
1616 # push projects but to enable mirror projects to be more easily converted to
1617 # push projects, we go ahead and do it for all projects.
1618 # By the time we get here we really shouldn't have any of these, but just in case.
1619 { find -L objects/$octet -type f -name "$octet19*" -exec chmod ug+w '{}' + || :; } 2>/dev/null
1621 # darcs mirrors have a xxx.log file that will grow endlessly
1622 # if this is a mirror and the file exists, shorten it to 10000 lines
1623 # also take this opportunity to optimize the darcs repo
1624 if ! [ -e .nofetch ] && [ -n "$cfg_mirror" ]; then
1625 url="$(config_get baseurl)" || :
1626 case "$url" in darcs://* | darcs+http://* | darcs+https://*)
1627 if [ -n "$cfg_mirror_darcs" ]; then
1628 url="${url%/}"
1629 basedarcs="$(basename "${url#darcs*:/}")"
1630 if [ -f "$basedarcs.log" ]; then
1631 tail -n 10000 "$basedarcs.log" >"$basedarcs.log.$$"
1632 mv -f "$basedarcs.log.$$" "$basedarcs.log"
1634 if [ -d "$basedarcs.darcs" ]; then
1636 cd "$basedarcs.darcs"
1637 # without show_progress suppress non-error output
1638 [ "${show_progress:-0}" != "0" ] || exec >/dev/null
1639 # Note that this does not optimize _darcs/inventories/ :(
1640 darcs optimize || :
1644 esac
1647 # Create a matching .bndl header file for the all-in-one pack we just created
1648 # but only if we're not a fork (otherwise the bundle would not be complete)
1649 # and we are running at least Git version 1.7.2 (pack_is_complete always fails otherwise)
1650 if [ -n "$makebndl" ] && [ -n "$var_have_git_172" ]; then
1651 # There should only be one pack in $packs but do some checking...
1652 # The one we just created will have a .idx and will NOT have a .keep
1653 progress "~ [$proj] creating downloadable bundle header"
1654 pkbase=
1655 pkhead=
1656 IFS= read -r curhead <repack/HEAD.orig || :
1658 [ -s "objects/pack/pack-$packs.pack" ] &&
1659 [ -s "objects/pack/pack-$packs.idx" ] &&
1660 ! [ -e "objects/pack/pack-$packs.keep" ] &&
1661 pkhead="$(pack_is_complete "$PWD/objects/pack/pack-$packs.pack" \
1662 "$PWD/repack/packed-refs.orig" "$curhead")"
1663 then
1664 pkbase="objects/pack/pack-$packs"
1666 if [ -n "$pkbase" ] && [ -n "$pkhead" ]; then
1668 symref=
1669 case "$curhead" in "ref: refs/"?*|"ref:refs/"?*|"refs/"?*)
1670 symref="${curhead#ref:}"
1671 symref="${symref# }"
1672 esac
1673 bndlurl=
1674 [ -z "$cfg_httpbundleurl" ] || bndlurl=" url=$cfg_httpbundleurl/$proj.git/clone.bundle"
1675 echo "# v2 git bundle"
1676 LC_ALL=C sed -ne "/^$octet20$hexdig* refs\/[^ $tab]*\$/ p" <repack/packed-refs.orig
1677 if [ -n "$symref" ]; then
1678 printf "$pkhead HEAD\0symref=HEAD:%s%s\n" "$symref" "$bndlurl"
1679 else
1680 if [ -n "$bndlurl" ]; then
1681 printf "$pkhead HEAD\0%s\n" "${bndlurl# }"
1682 else
1683 echo "$pkhead HEAD"
1686 echo ""
1687 } >"$pkbase.bndl"
1688 bndletag="$("$cfg_basedir/bin/rangecgi" --etag -m 1 "$pkbase.bndl" "$pkbase.pack")" || :
1689 bndlsha="$(printf '%s' "$bndletag" | git hash-object --stdin)" || :
1690 if [ -n "$bndletag" ]; then
1691 case "$bndlsha" in $octet20*)
1692 bndlshatrailer="${bndlsha#????????}"
1693 bndlshaprefix="${bndlsha%$bndlshatrailer}"
1694 bndlname="$(TZ=UTC date +%Y%m%d_%H%M%S)-${bndlshaprefix:-0}"
1695 [ -d bundles ] || mkdir bundles
1696 echo "${pkbase#objects/pack/}.bndl" >"bundles/$bndlname"
1697 echo "${pkbase#objects/pack/}.pack" >>"bundles/$bndlname"
1698 ln -s -f -n "$bndlname" bundles/latest
1699 esac
1704 # Record the size of this repo as the sum of its clone packed-refs + *.pack sizes as 1024-byte blocks
1705 eval "reposizek=$(( $(
1706 echo 0 $(du -k repack/packed-refs.orig $(printf 'objects/pack/pack-%s.pack ' $packs) 2>/dev/null |
1707 LC_ALL=C awk '{print $1}') |
1708 LC_ALL=C sed -e 's/ / + /g') ))"
1709 config_set_raw girocco.reposizek "${reposizek:-0}"
1711 read -r rsighash <repack/packed-refs.sighash || :
1713 # Now we're finally done with this
1714 rm -rf repack
1716 # Update various "signature" hashes used to help determine whether or
1717 # not it's okay to skip gc in the future (because there's nothing to do)
1719 psighash="$(if [ -n "$packs" ] || [ -n "$spacks" ]; then
1720 printf 'pack-%s.pack\n' $packs $spacks
1721 fi | LC_ALL=C sort | git hash-object -t blob --stdin)"
1722 config_set_raw girocco.gcsig.packs "$psighash"
1723 config_set_raw girocco.gcsig.refs "$rsighash"
1725 # We use $gcstart here to avoid a race where a push occurs during the gc itself
1726 # and the next future gc could be incorrectly skipped if we used the current
1727 # timestamp here instead
1728 config_set lastgc "$gcstart"
1729 rm -f "$lockf"
1731 progress "- [$proj] garbage check ($(date))"