3 # NOTE: additional options can be passed to git repack by specifying
4 # them after the project name, for example:
12 echo "Usage: gc.sh projname [extra-repack-args]" >&2
18 unset GIROCCO_SUPPRESS_AUTO_GC_UPDATE
19 .
"$cfg_basedir/jobd/maintain-auto-gc-hack.sh"
20 .
"$cfg_basedir/jobd/generate-auto-gc-update.sh"
21 GIROCCO_SUPPRESS_AUTO_GC_UPDATE
=1 && export GIROCCO_SUPPRESS_AUTO_GC_UPDATE
24 packopts
="--depth=50 --window=50 --window-memory=${var_window_memory:-1g}"
25 quiet
=; [ -n "$show_progress" ] || quiet
=-q
28 [ "$cfg_permission_control" != "Hooks" ] ||
umask 000
32 if _result
="$(kill -0 "$1" 2>&1)"; then
33 # process exists and we have permission to signal it
36 case "$_result" in *"not permitted"*)
37 # we do not have permission to signal the process
40 # process does not exist
45 # A .lock file should only exist for much less than a second.
46 # If we see a stale lock file (> 1h old), remove it and then,
47 # just in case, wait 30 seconds for any process whose .lock
48 # we might have just removed (it's racy) to finish doing what
49 # should take much less than a second to do.
50 _stalelock
="$(find -L "$1.lock
" -maxdepth 1 -mmin +60 -print 2>/dev/null)" ||
:
51 if [ -n "$_stalelock" ]; then
56 if (set -C; >"$1.lock") 2>/dev
/null
; then
61 [ "$_try" != "p" ] ||
sleep 1
63 # cannot create lock file
67 # The pre-receive script creates one ref log file per push but we want them to
68 # be coalesced into one ref log file per day. We are guaranteed that any files
69 # we find to coalesce are NOT currently being written to since they are always
70 # written first as temporary files and then moved into place. We attempt to
71 # transfer the most recent modification time to the coalesced log file which
72 # would step on its mod time if it were being written to directly, but if we
73 # find per-process ref log files then it must be a push project and the only
74 # thing that would write directly to the main per-day log file would be a
75 # mirror project so there's actually no conflict.
76 # Also, if the clock is wonky (or was futzed with) we may have both YYYYMMDD
77 # and YYYYMMDD.gz present in which case combine them into YYYYMMDD
79 [ -d reflogs
] ||
return 0
81 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 |
82 while read -r rname
; do
83 if [ -e "$rname.gz" ]; then
84 if [ -s "$rname" ]; then
85 # Presumably the .gz file must have been created before the non-gz
86 # file since it had to be uncompressed at some point therefore
87 # we need to append the non-gz contents to it but keep the non-gz
88 # contents timestamp so we rename to YYYYMMDD_ which will sort first
89 # and be picked up in the next step if we are interrupted in the middle.
90 # If a YYYYMMDD_ file already exists we append to it and transfer the
91 # timestamp. Finally we transfer the YYYYMMDD_ timestamp to the result
92 # and remove the YYYYMMDD_ temporary file leaving the result uncompressed.
93 if [ -e "${rname}_" ]; then
94 cat "$rname" >>"${rname}_"
95 touch -r "$rname" "${rname}_"
99 mv "$rname" "${rname}_"
101 gzip -d "$rname.gz" </dev
/null
102 [ -e "$rname" ] && ! [ -e "$rname.gz" ]
103 cat "${rname}_" >>"$rname"
104 touch -r "${rname}_" "$rname"
107 # Just remove the empty file to resolve the problem
112 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 |
113 while read -r rname
; do
114 logname
="${rname%%_*}"
115 # If someone's been futzing with the date, the file we want to
116 # append to could already have been compressed, so we just uncompress
117 # it here. The previous block guarantees we do not have both a compressed
118 # and uncompressed version present at the same time.
119 if [ -e "$logname.gz" ]; then
120 gzip -d "$logname.gz" </dev
/null
121 [ -e "$logname" ] && ! [ -e "$logname.gz" ]
123 cat "$rname" >>"$logname"
124 touch -r "$rname" "$logname"
126 if [ -e "$rname" ]; then
128 echo "! [$proj] failed to remove $rname" >&2
129 exit 1 # will only exit subshell created by "|"
135 # Remove any files in reflogs that are older than $cfg_reflogs_lifetime days
137 [ -d reflogs
] ||
return 0
138 exp
="$(( ${cfg_reflogs_lifetime:-1} * 1440 ))"
139 [ $exp -gt 0 ] || exp
=1440
140 [ $exp -le 43200 ] || exp
=43200
141 find -L reflogs
-maxdepth 1 -type f
-mmin "+$exp" -exec rm -f '{}' + ||
:
144 # Compact any reflogs that are not today's UTC date unless a .gz version exists
146 [ -d reflogs
] ||
return 0
147 _td
="reflogs/$(TZ=UTC date '+%Y%m%d')"
148 find -L reflogs
-maxdepth 1 -type f
-name "[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]" -print |
149 while read -r rname
; do
150 [ "$rname" != "$_td" ] ||
continue
151 ! [ -e "$rname.gz" ] ||
continue
152 gzip -9 "$rname" </dev
/null
156 # return true if there's more than one objects/pack-<sha>.pack file or
157 # ANY sha-1 files in objects
159 _packs
=$
(find -L objects
/pack
-type f
-name "pack-$octet20*.pack" -print |
head -n 2 | LC_ALL
=C
wc -l)
160 if [ $_packs != 1 ] && [ $_packs != 0 ]; then
163 _objs
=$
(find -L objects
/$octet -type f
-name "$octet19*" -print 2>/dev
/null |
head -n 1 | LC_ALL
=C
wc -l)
167 # make sure combine-packs uses the correct Git executable
168 run_combine_packs
() {
169 PATH
="$var_git_exec_path:$cfg_basedir/bin:$PATH" @basedir@
/jobd
/combine-packs.sh
"$@"
172 # combine the input pack(s) into a new pack (or possibly packs if packSizeLimit set)
173 # input pack names are read from standard input one per line delimited by the first
174 # ':', ' ' or '\n' character on the line (which allows gfi-packs to be read directly)
175 # all arguments, if any, are passed to pack-objects as additional options
176 # returns non-zero on failure AND creates .gc_failed in that case
179 find -L objects
/pack
-maxdepth 1 -type f
-name '*.zap*' -exec rm -f '{}' + ||
:
180 run_combine_packs
--replace "$@" $packopts --all-progress-implied $quiet --non-empty ||
{
187 # if the current directory is_gfi_mirror then repack all packs listed in gfi-packs
189 [ -n "$gfi_mirror" ] ||
return 0
190 [ -d objects
/pack
] ||
{ rm -f gfi-packs
; return 0; }
191 progress
"~ [$proj] redeltifying poor quality git fast-import packs"
192 combine_packs
--ignore-missing --no-reuse-delta <gfi-packs
197 # combine small packs into larger pack(s)
198 # we avoid any keep, bndl or bitmap packs
199 # if the optional argument is non-empty even a single small pack will be redeltad
200 combine_small_packs
() {
203 if [ -n "$1" ] && [ -n "$noreusedeltaopt" ]; then
206 _lpo
="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl"
207 _lpo
="$_lpo --exclude-sfx _u --exclude-sfx _o"
208 _lpo
="$_lpo --quiet --object-limit $var_redelta_threshold objects/pack"
210 _cnt
="$(list_packs --count $_lpo)" ||
:
211 test "${_cnt:-0}" -ge $_minsmallpacks
213 [ -n "$_didprogress" ] ||
{
214 progress
"~ [$proj] combining small packs into a single larger pack"
217 _newp
="$(list_packs $_lpo | combine_packs --names $noreusedeltaopt)"
218 _newc
="$(( $(echo "$_newp" | LC_ALL=C wc -w) ))"
219 # be paranoid and exit the loop if we haven't reduced the number of packs
220 [ $_newc -lt $_cnt ] ||
break
226 # Unfortunately git-svn lacks the ability to store newly fetched revisions as a pack.
227 # However, the fetch code conveniently sets .svnpack just before it runs git-svn fetch
228 # so that it's easy to find all the objects that have been fetched by git-svn and
229 # combine them into a pack. The --no-reuse-delta option is meaningless here since
230 # everything to be packed is a loose object and therefore not a delta so deltification
231 # will always take place.
233 [ -f .svnpack
] && [ -n "$svn_mirror" ] ||
return 0
235 mv -f .svnpack .svnpackgc
236 progress
"~ [$proj] combining loose git-svn objects into a pack"
237 _newp
="$(find -L objects/$octet -maxdepth 1 -type f -newer .svnpackgc -name "$octet19*" -print 2>/dev/null |
238 LC_ALL=C awk -F / '{print $2 $3}' |
239 run_combine_packs --objects --names $packopts --incremental --all-progress-implied $quiet --non-empty)" ||
{
240 mv -f .svnpackgc .svnpack
244 if [ -n "$_newp" ]; then
245 # remove the now-redundant loose objects -- this is always safe
246 # even during a concurrent push because a reprepare_packed_git
247 # will be triggered if an object that should be there is not
248 # found thereby finding it in the new pack instead
249 git prune-packed
$quiet
254 # HEADSHA="$(pack_is_complete /full/path/to/some.pack /full/path/to/packed-refs "$(cat HEAD)")"
256 # Must have a matching .idx file and a non-empty packed-refs file
257 [ -s "${1%.pack}.idx" ] ||
return 1
258 [ -s "$2" ] ||
return 1
264 "ref: refs/"?
*|
"ref:refs/"?
*|
"refs/"?
*)
265 _headmatch
="${3#ref:}"
266 _headmatch
="${_headmatch# }"
267 _headmatchpat
="$(echo "$_headmatch" | LC_ALL=C sed -e 's/\([.$]\)/\\\1/g')"
268 _headsha
="$(LC_ALL=C grep -e "^
$octet20$hexdig* $_headmatchpat\$
" <"$2" |
269 LC_ALL=C cut -d ' ' -f 1)"
270 case "$_headsha" in $octet20*) :;; *)
278 rm -rf pack_is_complete_test
279 mkdir pack_is_complete_test
280 mkdir pack_is_complete_test
/refs
281 mkdir pack_is_complete_test
/objects
282 mkdir pack_is_complete_test
/objects
/pack
283 echo "$_headsha" >pack_is_complete_test
/HEAD
284 ln -s "$1" pack_is_complete_test
/objects
/pack
/
285 ln -s "${1%.pack}.idx" pack_is_complete_test
/objects
/pack
/
286 ln -s "$2" pack_is_complete_test
/packed-refs
287 _count
="$(git --git-dir=pack_is_complete_test rev-list --count --all 2>/dev/null)" ||
:
288 rm -rf pack_is_complete_test
289 [ -n "$_count" ] ||
return 1
290 [ "$_count" -gt 0 ] 2>/dev
/null ||
return 1
294 # On return a "$lockf" will have been created that must be removed when gc is done
296 # be compatibile with gc.pid file from newer Git releases
300 if [ "$(createlock "$lockf")" ]; then
302 # 1) less than 12 hours old
303 # 2) contains two fields (pid hostname) NO trailing NL
304 # 3) the hostname is different OR the pid is still alive
305 # then we exit as another active process is holding the lock
306 if [ "$(find -L "$lockf" -maxdepth 1 -mmin -720 -print 2>/dev/null)" ]; then
309 read -r apid ahost ajunk
<"$lockf" ||
:
310 if [ "$apid" ] && [ "$ahost" ]; then
311 if [ "$ahost" != "$hn" ] || pidactive
"$apid"; then
317 echo >&2 "[$proj] unable to create gc.pid.lock file"
320 if [ -n "$active" ]; then
322 echo >&2 "[$proj] gc already running on machine '$ahost' pid '$apid'"
325 printf "%s %s" "$$" "$hn" >"$lockf.lock"
326 chmod 0664 "$lockf.lock"
327 mv -f "$lockf.lock" "$lockf"
330 # Create a repack subdirectory such that running repack in it will pack the
331 # same things that a pack in the normal directory would except that the pack
332 # is guaranteed to be generated in an optimized order by adding a suitable
333 # synthesized ref in the refs/tags namespace (yes, pack-objects.c really does
334 # behave differently depending on the contents of the refs/tags namespace).
335 # Before calling this, pack-refs --all MUST be performed or the wrong pack
336 # will end up being made.
338 # If a ref deletion is pushed after making the repack subdir but before the
339 # the actual repack, the discarded objects will be packed -- no big deal,
340 # they'll get discarded the next time gc runs.
342 # If a fast-forward ref update is pushed after making the repack subdir but
343 # before the actual repack, it will be picked up and the new objects packed
344 # (subject to the normal git repack race about picking such updates up).
346 # If a non-fast-forward ref update is pushed after making the repack subdir but
347 # before the actual repack, it will be picked up like a fast-forward update but
348 # the discarded objects will be included like a ref deletion (until the next
349 # scheduled gc takes place).
351 # We retain a copy of the original packed-refs file as repack/packed-refs.orig
352 # If ref deletions come in while we're repacking, the original packed-refs
353 # file will be modified, but we'll still pack the deleted ref(s).
354 # If the packed-refs.orig file is used to create the bundle header we avoid
355 # a situation where the bundle contains a ref state that never actually
356 # existed in reality (for example a new branch is pushed and then an old
357 # branch deleted afterwards -- the deletion would show up in the bundle
358 # because it will cause the original packed-refs file to be re-written, but
359 # the new branch creation will not unless we do another pack-refs which might
360 # lead to having in incomplete bundle). Therefore we want to keep a copy of
361 # the original packed-refs file around. We do the same thing for HEAD.
363 # It's possible that the "objects" subdirectory is a symbolic link.
364 # Git does support this. However, during the repacking process, new packs
365 # will be created in repack/alt/pack and then moved into objects/pack.
366 # In order for this to work seemlessly, they must both be on the same
367 # filesystem. But when objects (or even objects/pack) is a symbolic link they
368 # might not be. For this reason a "repack" subdirectory is created under
369 # objects/pack and the repack/alt/pack directory symbolicly linked to it.
371 # Git allows not just HEAD to be a symbolic-ref, but any ref anywhere in the
372 # refs namespace. We are concerned about ref name collisions and getting the
373 # right tag set to get an optimal pack. We can safely duplicate the ref space
374 # under refs/heads, refs/notes and refs/remotes without any risk of unwanted
375 # collisions and this will likely make over 99%+ of all symbolic refs found
376 # in the wild work properly. Girocco itself never creates any symbolic refs
377 # inside the refs namespace; this is a nod to simultaneously using a Girocco
378 # repository for other purposes.
380 ! [ -d repack
] ||
rm -rf repack
381 ! [ -d repack
] ||
{ echo >&2 "[$proj] cannot remove repack subdirectory"; exit 1; }
382 [ -d objects
/pack
] || mkdir
-p objects
/pack
383 ! [ -d objects
/pack
/repack
] ||
rm -rf objects
/pack
/repack
384 ! [ -d objects
/pack
/repack
] ||
{ echo >&2 "[$proj] cannot remove objects/pack/repack subdirectory"; exit 1; }
385 mkdir repack repack
/refs repack
/alt objects
/pack
/repack
386 [ -d info
] || mkdir info
387 ln -s ..
/config repack
/config
388 ln -s ..
/info repack
/info
389 ln -s ..
/objects repack
/objects
390 ln -s "$PWD/objects/pack/repack" repack
/alt
/pack
391 ln -s ..
/..
/refs repack
/refs
/refs
392 _lines
=$
(( $
(LC_ALL
=C
wc -l <packed-refs
) ))
393 cat HEAD
>repack
/HEAD.orig
394 cat packed-refs
>repack
/packed-refs.orig
395 if [ $
(LC_ALL
=C
wc -l <repack
/packed-refs.orig
) -ne "$_lines" ]; then
396 echo >&2 "[$proj] error: make_repack_dir failed original packed-refs line count sanity check"
399 # Note: Git v1.5.0 introduced the "# pack-refs with:" header line for the packed-refs file
400 sed '/^# pack-refs/d; s, refs/, refs/!/,' <repack
/packed-refs.orig
>repack
/packed-refs
401 headref
="$(git rev-parse --verify --quiet HEAD)" ||
:
402 if [ -n "$headref" ]; then
403 echo "$headref refs/!=/HEAD" >>repack
/packed-refs
404 echo "$headref refs/heads/!" >>repack
/packed-refs
405 _lines
=$
(( $_lines + 2 ))
407 if [ $
(( $
(LC_ALL
=C
wc -l <repack
/packed-refs
) + 1 )) -ne "$_lines" ]; then
408 echo >&2 "[$proj] error: make_repack_dir failed packed-refs initial line count sanity check"
411 sed -n '\, refs/heads/,p; \, refs/notes/,p; \, refs/remotes/,p' <repack
/packed-refs.orig
>>repack
/packed-refs
412 _newlines
="$(( $(LC_ALL=C wc -l <repack/packed-refs) ))"
413 if [ $
(( $_newlines + 1 )) -lt "$_lines" ]; then
414 echo >&2 "[$proj] error: make_repack_dir failed packed-refs extra line count sanity check"
418 optref
="$(git rev-list -n 1 --all 2>/dev/null)" ||
:
419 if [ -n "$optref" ]; then
420 echo "$optref refs/tags/!" >>repack
/packed-refs
421 _lines
=$
(( $_lines + 1 ))
422 echo "$optref" >repack
/HEAD
424 cat HEAD
>repack
/HEAD
426 if [ $
(LC_ALL
=C
wc -l <repack
/packed-refs
) -ne "$_lines" ]; then
427 echo >&2 "[$proj] error: make_repack_dir failed packed-refs line count sanity check"
432 # Remove any crud that's been left behind by interrupted operations
433 # that did not clean up after themselves
435 # Remove any existing FETCH_HEAD
436 # There can only be a FETCH_HEAD if we've been fetching, not if we've been
437 # receiving pushes (those never create a FETCH_HEAD).
438 # And if we're fetching because we're a mirror, we know we're not fetching right
439 # now since jobd.pl never runs a project's fetch simultaneously with its gc.
440 # Therefore any existing FETCH_HEAD is junk. And it may be many megabytes if
441 # there were a lot of refs.
444 # remove any existing pack_is_complete_test or repack subdirectories
445 # If either exists when this function is called it's crud
446 rm -rf pack_is_complete_test repack objects
/pack
/repack
448 # Remove any stale pack remnants that are more than an hour old.
449 # Stale pack fragments are defined as any pack-<sha1>.ext where .ext is NOT
450 # .pack AND the corresponding .pack DOES NOT exist. A bunch of stale
451 # pack-<sha1>.idx files without their corresponding .pack files are worthless
452 # and just waste space. Normally there shouldn't be any remnants but actually
453 # this can happen when things are interrupted at just the wrong time.
454 # Note that the objects/pack directory is created by git init and should
456 find -L objects
/pack
-maxdepth 1 -type f
-mmin +60 -name "pack-$octet20*.?*" -print |
457 LC_ALL
=C
sed -e 's/^objects\/pack\/pack-//; s/\..*$//' | LC_ALL
=C
sort -u |
458 while read packsha
; do
459 ! [ -e "objects/pack/pack-$packsha.pack" ] ||
continue
460 rm -f "objects/pack/pack-$packsha".?
*
463 # Remove any stale tmp reflogs files that are more than one hour old.
464 # Since they are created only while the pre-receive hook is running and
465 # all it does is process a bunch of refs passed to it on standard input
466 # it's inconceivable that it would ever take as much as an hour to run.
467 if [ -d reflogs
]; then
468 find -L reflogs
-maxdepth 1 -type f
-mmin +60 -name "tmp_*" -exec rm -f '{}' + ||
:
471 # Remove any stale object tmp_obj_* files that are more than 3 hours old.
472 # Really these files should only exist very briefly so there shouldn't be any
473 # but things happen that can end up leaving them behind.
474 find -L objects
/$octet -maxdepth 1 -type f
-mmin +180 -name "tmp_obj_?*" -exec rm -f '{}' + 2>/dev
/null ||
:
476 # Remove any stale pack .keep files that are more than 12 hours old.
477 # We don't do anything to create any permanent pack .keep files, so they must
478 # be remnants from some failed push or something. Removing the .keep will
479 # allow the pack to be properly repacked.
480 find -L objects
/pack
-maxdepth 1 -type f
-mmin +720 -name "pack-$octet20*.keep" -exec rm -f '{}' + ||
:
482 # Remove any stale tmp_pack_*, tmp_idx_*, tmp_bitmap_*, packtmp-* or .tmp-*-pack* files
483 # that are more than 12 hours old.
484 find -L objects
/pack
-maxdepth 1 -type f
-mmin +720 \
( \
485 -name "tmp_pack_?*" -o -name "tmp_idx_?*" -o -name "tmp_bitmap_?*" -o \
486 -name "packtmp-?*" -o -name ".tmp-?*-pack*" \
487 \
) -exec rm -f '{}' + ||
:
489 # Remove any stale incoming-* object quarantine directories that are
490 # more than 12 hours old. These are new with Git >= 2.11.0.
491 find -L objects
-maxdepth 1 -type d
-name 'incoming-?*' -mmin +720 \
492 -exec rm -rf '{}' + ||
:
494 # Remove any stale shallow_* files that are more than 12 hours old.
495 # These can be left behind by Git >= 1.8.4.2 and < 2.0.0 when a client
496 # requests a shallow clone. Also discard stale .refs-temp* and
497 # .refs-new* files at the same time.
498 find -L .
-maxdepth 1 -type f
-mmin +720 \
( \
499 -name "shallow_?*" -o -name ".refs-temp*" -o -name ".refs-new*" \
500 \
) -exec rm -f '{}' + ||
:
502 # Remove any stale *.temp files in the objects area that are more than 12 hours old.
503 # This can be stale sha1.temp, or stale *.pack.temp so we kill all stale *.temp.
504 find -L objects
-type f
-mmin +720 -name "*.temp" -exec rm -f '{}' + ||
:
506 # Remove any stale *.lock files in the htmlcache area that might have been left
507 # behind after an abnormal exit during an attempt to update a cached file and
508 # are more than 1 hour old.
509 ! [ -d htmlcache
] ||
find -L htmlcache
-type f
-mmin +60 -name "*.lock" -exec rm -f '{}' + ||
:
511 # Remove any stale git-svn temp files that are more than 12 hours old.
512 # The git-svn process creates temp files with random 10 character names
513 # in the root of $GIT_DIR. Unfortunately they do not have a recognizable
514 # prefix, so we just have to kill any files with a 10-character name. We
515 # do this only for git-svn mirrors. All characters are chosen from
516 # [A-Za-z0-9_] so we can at least check that and fortunately the only
517 # collision is 'FETCH_HEAD' but that shouldn't matter.
518 # There may also be temp files with a Git_ prefix as well.
519 if [ -n "$svn_mirror" ]; then
520 _randchar
='[A-Za-z0-9_]'
521 _randchar2
="$_randchar$_randchar"
522 _randchar4
="$_randchar2$_randchar2"
523 _randchar10
="$_randchar4$_randchar4$_randchar2"
524 find -L .
-maxdepth 1 -type f
-mmin +720 -name "$_randchar10" -exec rm -f '{}' + ||
:
525 find -L .
-maxdepth 1 -type f
-mmin +720 -name "Git_*" -exec rm -f '{}' + ||
:
528 # Remove any stale fast_import_crash_<pid> files that are more than 3 days old.
529 if [ -n "$gfi_mirror" ]; then
530 find -L .
-maxdepth 1 -type f
-mmin +4320 -name "fast_import_crash_?*" -exec rm -f '{}' + ||
:
533 # Remove any stale core or *.core or core.* files that are more than 3 days old.
534 find -L .
-maxdepth 1 -type f
-mmin +4320 \
( -name "core" -o -name "*.core" -o -name "core.*" \
) \
535 -exec rm -f '{}' + ||
:
539 ## Garbage Collection Types
541 ## There are two kinds of possible garbage collection (gc) operations:
543 ## 1. A normal, full gc
546 ## If the full garbage collection interval has expired (or gc has never been
547 ## run), then a normal, full gc will take place. Otherwise, a "mini" gc will
548 ## take place if the file .needsgc exists.
550 ## A "mini" gc is similar to "git gc --auto" in that it may not end up actually
551 ## doing anything unless the right conditions are present so it's not a burden
552 ## to run it often. If the file .needsgc exists, a "mini" gc will occur at
553 ## the next opportunity.
555 ## Note, however, that the .nogc file suppresses ALL gc activity (normal or mini).
560 cd "$cfg_reporoot/$proj.git"
561 [ -d objects
/pack
] ||
{ rm -f gfi-packs
; mkdir
-p objects
/pack
; }
562 mirror_url
="$(get_mirror_url)" ||
:
564 ! is_svn_mirror_url
"$mirror_url" || svn_mirror
=1
566 if [ -f gfi-packs
] && [ -s gfi-packs
] && is_gfi_mirror_url
"$mirror_url"; then
570 # If git config --bool --get girocco.redelta is explicitly false then automatic
571 # redelta when there are less than $var_redelta_threshold objects will be suppressed.
572 # On the other hand, if git config --get girocco.redelta is "always" then, on a full
573 # gc only, for the final repack, deltas will always be recomputed.
574 # This can be set on a per-project basis to avoid unusual pathological gc behavior.
575 # Setting this will hurt efficiency of the affected repository.
576 # Note that fast-import packs ALWAYS get new deltas regardless of this setting.
577 noreusedeltaopt
="--no-reuse-delta"
578 [ "$(git config --bool --get girocco.redelta 2>/dev/null || :)" != "false" ] || noreusedeltaopt
=
580 [ "$(git config --get girocco.redelta 2>/dev/null || :)" != "always" ] || alwaysredelta
=1
582 # Extract any -f or -F or --no-reuse-object or --no-reuse-delta options
583 # to be compatible with the old and new gc.sh versions and avoid ugly argument
584 # duplication in process lists at the same time
585 # Any options found will override the "girocco.redelta" setting
588 while [ $idx -gt 0 ]; do
597 -F|
--no-reuse-object)
605 printf >&2 '%s\n' "bad non-option argument: $opt"
606 echo >&2 "(Did you perhaps intend to use a --xxx=yyy form?)"
609 [ -z "$opt" ] ||
set -- "$@" "$opt"
611 if [ -n "$alwaysredelta" ]; then
612 noreusedeltaopt
="--no-reuse-delta"
613 [ -z "$recompress" ] || noreusedeltaopt
="--no-reuse-object"
616 trap 'e=$?; rm -f .gc_in_progress; if [ $e != 0 ]; then echo "gc failed dir: $PWD" >&2; fi' EXIT
620 # date -R is linux-only, POSIX equivalent is '+%a, %d %b %Y %T %z'
621 datefmt
='+%a, %d %b %Y %T %z'
624 if [ "${force_gc:-0}" = "0" ] && check_interval lastgc
$cfg_min_gc_interval; then
625 if [ -e .needsgc
]; then
628 progress
"= [$proj] garbage check skip (last at $(config_get lastgc))"
632 if [ -e .nogc
]; then
633 progress
"x [$proj] garbage check disabled"
636 if [ -z "$isminigc" ] && [ -e .delaygc
] && [ -e .needsgc
]; then
637 # Eligible for a full gc but .delaygc is set so it would be skipped
638 # However .needsgc is also set so transform it into a mini instead
640 progress
"~ [$proj] garbage check delayed but checking mini because .needsgc"
643 if [ -n "$isminigc" ]; then
644 # Perform a "mini" gc
645 # Note that .delaygc is ignored here as that's only intended for full gc
647 rm -f .allowgc .needsgc
648 rm -f objects
/pack
/pack-
*_r.keep
653 maintain_auto_gc_hack
654 generate_auto_gc_update
656 if [ -f .svnpack
] && [ -n "$svn_mirror" ]; then
658 progress
"+ [$proj] mini garbage check ($(date))"
661 if [ -z "$cfg_delay_gfi_redelta" ] && [ -n "$gfi_mirror" ]; then
662 # $Girocco::Config::delay_gfi_redelta is false, force redeltification now
663 if [ -z "$miniactive" ]; then
665 progress
"+ [$proj] mini garbage check ($(date))"
669 # If there aren't at least 10 non-keep, non-bitmap, non-bndl packs then
670 # don't actually process them yet
671 lpo
="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl --quiet"
672 packcnt
="$(list_packs --count $lpo objects/pack)" ||
:
673 if [ "${packcnt:-0}" -ge 10 ]; then
674 if [ -z "$miniactive" ]; then
676 progress
"+ [$proj] mini garbage check ($(date))"
678 if [ -n "$gfi_mirror" ]; then
680 packcnt
="$(list_packs --count $lpo objects/pack)" ||
:
682 # if repack_gfi_packs dropped the pack count to < 10 don't combine
683 if [ "${packcnt:-0}" -ge 10 ]; then
685 packcnt
="$(list_packs --count $lpo objects/pack)" ||
:
687 # if we still have more than 10 packs trigger a full gc
688 if [ "${packcnt:-0}" -ge 10 ]; then
689 # We shouldn't be in a .delaygc state at this point, but if
690 # we are then nuke it because we really need a full gc now
692 git config
--unset gitweb.lastgc
694 git update-server-info
# just in case
695 progress
"- [$proj] mini garbage check triggering full gc too many packs ($(date))"
700 if [ -n "$miniactive" ]; then
701 git update-server-info
702 progress
"- [$proj] mini garbage check ($(date))"
704 progress
"= [$proj] mini garbage check nothing but crud removal to do ($(date))"
709 # Avoid unnecessary garbage collections:
710 # 1. If lastreceive is set and is older than lastgc
712 # 2. We are not a fork (! -s alternates) -OR- lastparentgc is older than lastgc
714 # If lastgc is NOT set or lastreceive is NOT set we MUST run gc
715 # If we are a fork and lastparentgc is NOT set we MUST run gc
717 # If the repo is dirty after removing any crud we MUST run gc
719 gcstart
="$(date "$datefmt")"
722 ! [ -s objects
/info
/alternates
] || isfork
=1
724 [ -z "$isfork" ] || lastparentgcsecs
="$(config_get_date_seconds lastparentgc)" ||
:
726 if lastreceivesecs
="$(config_get_date_seconds lastreceive)" &&
727 [ "${force_gc:-0}" = "0" ] &&
728 lastgcsecs
="$(config_get_date_seconds lastgc)" &&
729 [ $lastreceivesecs -lt $lastgcsecs ]; then
730 # We've run gc since we last received, so maybe we can skip,
731 # check if not fork or fork and lastparentgc < lastgc
732 if [ -n "$isfork" ]; then
733 if [ -n "$lastparentgcsecs" ] &&
734 [ $lastparentgcsecs -lt $lastgcsecs ]; then
735 # We've run gc since our parent ran gc so we can skip
739 # We don't have any alternates (we're not a forK) so we can skip
744 # Prevent any other simultaneous gc operations
747 # At this point, if .allowgc or .gc_failed exists, it's now crud to be removed
748 rm -f .allowgc .gc_failed
750 # Ideally we would do this in post-receive, but that would mean duplicating the
751 # logic so it's available in the chroot jail and that's highly undesirable
752 # Instead, since the first gc will be triggered immediately following the first
753 # push, we do the check here as it's quick and harmless if HEAD is already valid
754 check_and_set_head ||
:
756 # Always get rid of crud
759 # Always perform reflogs maintenance
764 # Always maintain auto gc hack
765 maintain_auto_gc_hack
766 generate_auto_gc_update
768 # Run 'git svn gc' now for svn mirrors
769 if [ -n "$svn_mirror" ]; then
773 # Skip the actual gc if .delaygc is set
774 if [ -e .delaygc
]; then
775 progress
"x [$proj] garbage check delayed (except for crud removal)"
780 # Do not skip gc if the repo is dirty
781 if [ -n "$skipgc" ] && ! is_dirty
; then
782 progress
"= [$proj] garbage check nothing but crud removal to do ($(date))"
783 config_set lastgc
"$gcstart"
789 if [ -n "$isfork" ] && [ -z "$lastparentgcsecs" ]; then
790 # set lastparentgc and then update gcstart to be at least 1 second later
791 config_set lastparentgc
"$gcstart"
794 if [ -z "$lastreceivesecs" ]; then
795 # set lastreceive and then update gcstart to be at least 1 second later
796 config_set lastreceive
"$gcstart"
799 if [ -n "$bumptime" ]; then
801 gcstart
="$(date "$datefmt")"
804 progress
"+ [$proj] garbage check ($(date))"
807 [ -z "$alwaysredelta" ] || newdeltas
="$noreusedeltaopt"
808 if [ -z "$newdeltas" ] && [ -n "$gfi_mirror" ]; then
809 if [ $
(list_packs
--exclude-no-idx --count objects
/pack
) -le \
810 $
(list_packs
--exclude-no-idx --count --quiet --only gfi-packs
) ]; then
811 # Don't bother with repack_gfi_packs since everything's being repacked
812 newdeltas
="--no-reuse-delta"
815 if [ -z "$newdeltas" ] && [ -n "$noreusedeltaopt" ] &&
816 [ $
(list_packs
--exclude-no-idx --count-objects objects
/pack
) -le $var_redelta_threshold ]; then
817 # There aren't enough objects to worry about so just redelta to get the best pack
818 newdeltas
="--no-reuse-delta"
820 if [ -z "$newdeltas" ]; then
821 # Since we're not going to recompute deltas overall, we need to do the
822 # "mini" maintenance so that we can get more optimal deltas
823 [ -z "$noreusedeltaopt" ] || make_svn_pack
825 force_single_pack_redelta
=
826 [ -n "$gfi_mirror" ] ||
[ -n "$svn_mirror" ] || force_single_pack_redelta
=1
827 [ -z "$noreusedeltaopt" ] || combine_small_packs
$force_single_pack_redelta
831 ## Safe Pruning In Forks
833 ## We are about to perform garbage collection. We do NOT use the "git gc" or
834 ## the "git repack" commands directly as they do not provide enough control over
835 ## the fine details. However, we DO maintain a "gc.pid" file during our garbage
836 ## collection so that a simultaneous "git gc" by an administrator will be
837 ## blocked (and similarly we refuse to start garbage collection if we cannot
838 ## create the "gc.pid" file).
840 ## When we say "gc" in the below description we are referring to our "gc.sh"
841 ## script, NOT the "git gc" command.
843 ## If the project we are running garbage collection (gc) on has any forks we
844 ## must be careful not to remove any objects that while no longer referenced by
845 ## this project (the parent) are still referenced by one or more forks (the
846 ## children) otherwise the children will become corrupt and we can't abide
849 ## One way to accomplish this is to simply hard-link all currently existing
850 ## loose objects and packs in the parent into all the children that refer to the
851 ## parent (via a line in their objects/info/alternates file) before beginning
852 ## the gc operation and then relying on a subsequent gc in the child to clean up
853 ## any excess objects/packs. We used to use this strategy but it's very
854 ## inefficient because:
856 ## 1. The disk space used by the old pack(s)/object(s) will not be reclaimed
857 ## until all children (and their children, if any) run gc by which time
858 ## it's quite possible the topmost parent will have run gc again and
859 ## hard-linked yet another old pack down to its children (not to mention
862 ## 2. When using the "-A" option with "git repack", any new objects in the
863 ## parent that are not referenced by children will continually get
864 ## exploded out of the hard-linked pack in the children whenever the
867 ## 3. To avoid suboptimal and/or unnecessarily many packs being hard-linked
868 ## into child forks, we must run the "mini" gc maintenance before we
869 ## perform the hard-linking into the children which provides yet another
870 ## source of inefficiency.
872 ## While we were still using the "-A" option to "git repack" (that was not
873 ## always the case) to guarantee we can access old ref values for long enough
874 ## to send out a meaningful mail.sh notification, another, more efficient,
875 ## option became available to prevent corruption of child forks that continue
876 ## to refer to objects that are no longer reachable from any ref in the parent.
878 ## The only things that need be copied (or hard-linked) into the child fork(s)
879 ## are those objects that have become unreachable from any ref in the parent.
881 ## When we were using the "git repack -A -d" + "git prune --expire=1.day.ago"
882 ## technique, the only objects that could ever be removed were loose objects
883 ## that "git prune" determined were expired. In that case, loose objects were
884 ## all that need be hard-linked down to child forks in order to avoid
885 ## corruption of any child fork(s).
887 ## The "git repack -A -d" + "git prune --expire=1.day.ago" + hard-linking loose
888 ## objects to child forks technique remains fundamentally sound from the
889 ## perspective of supporting simultaneous gc and push and keeping newly
890 ## unreachable objects around long enough to be sure we can send out meaningful
891 ## ref change notifications and never corrupting any child forks and never
892 ## persisting the lifetime of large old packs containing mostly duplicate or
893 ## unreachable objects as gc percolates through a project's entire fork tree.
895 ## However, that technique suffers from one potential prodigious pitfall.
897 ## Unreachable objects come flying out of their packs to splatter all over the
898 ## objects subdirectories possibly creating a huge, inefficient mess.
900 ## Often this is not an issue. Even with a lot of rebasing going on, usually
901 ## the only objects that will splatter are some commits, trees and the odd blob
902 ## here and there. Not enough to be overly concerned about.
904 ## However, for the reppository that frequently experiences a lot of non-fast-
905 ## forward updates and/or outright ref deletion, the number of objects suddenly
906 ## popping out of their packs at "git repack -A -d" time can be overwhelming.
908 ## To avoid this issue we now use a four phase pack creation strategy.
909 ## This will result in creation of up to four packs (instead of at most one).
911 ## I. A complete pack (with bitmaps if appropriate) gets created including
912 ## only "reachable" objects from all refs/... refs plus HEAD. This will
913 ## also serve as the virtual bundle for the repository.
915 ## II. A pack of recently-became-unreachable objects and friends is created.
916 ## (The "friends" are ref logs, linked working tree HEADs and indicies.)
917 ## Because both the pre-receive and update.sh script record all ref
918 ## changes we can easily choose the cut off point for "recently".
919 ## It is only the fact we maintain those logs in the reflogs subdirectory
920 ## that allows this step to be possible.
922 ## III. If the repository has any forks with a non-zero length alternates file,
923 ## yet another pack of "--keep-unreachable" objects is generated that will
924 ## not actually be kept in the parent, but hard-linked into all the forks.
926 ## IV. Finally, after running "git prune-packed", any remaining loose objects
927 ## are migrated into a pack of their own.
929 ## We then remove any non-.keep packs that existed before we started the
930 ## process being careful to keep any same-pack pushes for the "Push Pack Redux"
931 ## race condition (see README-GC).
933 ## By using "git pack-objects" directly we are able to accomplish this with
934 ## very little additional effort.
936 ## The packs produced by (III) are treated almost like ".keep" packs by child
937 ## forks in that the objects in them are never repacked into any other
938 ## "--keep-unreachable" packs (but they can migrate into phase I or II packs)
939 ## and those phase III packs are then hard-linked into any grandchild forks.
941 ## This avoids the space explosion that could occur if each fork level ended
942 ## up duplicating the "--keep-unreachable" pack space by repacking those
943 ## objects (essentially breaking the hard-link to the single copy of those
946 ## While it is true that each level of forks could potentially add yet another
947 ## phase III pack to be hard-linked down to its children, such packs will only
948 ## include unreachable objects not already in any phase III packs that were
949 ## received from the parent.
951 ## The space for the phase III packs will not be reclaimed until the gc
952 ## finishes percolating through the entire "fork tree" of a project.
954 ## This is not much different than the "git repack -A -d" situation where
955 ## all the loose objects are hard-linked down into child forks. In that
956 ## case forks that actually need any of those objects could gradually reduce
957 ## the number of objects hard-linked into deeper fork levels.
959 ## The difference with a phase III "--keep-unreachable" pack is that there
960 ## cannot be any gradual reduction like that since it would require repacking
961 ## the pack and breaking the hard-link thereby increasing storage space. The
962 ## storage will instead always be reclaimed all at once when all of the
963 ## projects in the "fork tree" complete their gc.
965 ## However, the belief is that the huge space win by having all the
966 ## unreachable objects packed up together far eclipses (when many objects are
967 ## involved, the single-pack version can end up using 1/20th or less of the
968 ## disk space compared to having them all as loose objects) any brief minor
969 ## space savings that might occur under the "git repack -A -d" loose object
970 ## system prior to the gc collection completing for all the projects in the
978 # rename_pack oldnamepath newnamepath
979 # note that .keep files are left untouched and not moved at all!
981 [ $# -eq 2 ] && [ "$1" != "$2" ] ||
{
982 echo >&2 "[$proj] incorrect use of rename_pack function"
985 # Git assumes that if the destination of the rename already exists
986 # that it is, in fact, a copy of the same bytes so silently succeeds
987 # without doing anything. We duplicate that logic here.
988 # Git checks for the .idx file first before even trying to use a pack
989 # so it should be the last moved and the first removed.
990 for ext
in pack bitmap idx
; do
991 [ -f "$1.$ext" ] ||
continue
992 ln "$1.$ext" "$2.$ext" >/dev
/null
2>&1 ||
993 [ -f "$2.$ext" ] ||
{
994 echo >&2 "[$proj] unable to move $1.$ext to $2.$ext"
998 for ext
in idx pack bitmap
; do
1005 find -L "$1" -maxdepth 1 -type f
! -perm -ug+w \
1006 -name "pack-$octet20*.pack" -exec chmod ug
+w
'{}' + ||
:
1010 eval "$1="'$(( $# - 1 ))'
1014 if [ -s "$1" ]; then
1016 export GIT_INDEX_FILE
1017 git write-tree
2>/dev
/null ||
:
1018 unset GIT_INDEX_FILE
1022 get_detached_head
() {
1023 if [ -s "$1" ] && read -r _head
<"$1" 2>/dev
/null
; then
1024 case "$_head" in $octet20*)
1030 # compute_extra_reachables
1031 # create lines suitable for a packed-refs file mentioning all the
1032 # other refs we might like to keep.
1033 # the current directory MUST be set to the repository's --git-dir
1034 # the following are included:
1035 # * refs mentioned in reflogs/... files
1036 # * tree(s) created from index file(s)
1037 # * detached linked working tree heads
1038 # Resulting objects are tested for existence and uniqified then output
1039 # one per line under a refs/z* namespace
1040 compute_extra_reachables
() {
1042 digits8
='[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]'
1043 find -L reflogs
-mindepth 1 -maxdepth 1 -type f
-name "$digits8*" -exec gzip -c -d -f '{}' + |
1044 awk '{print $2; print $3}'
1045 ! [ -f index
] || get_index_tree index
1046 if [ -d worktrees
]; then
1047 find -L worktrees
-mindepth 2 -maxdepth 2 -name HEAD
-type f
-print |
1048 while read -r lwth
; do
1049 get_detached_head
"$lwth"
1050 get_index_tree
"${lwth%HEAD}index"
1053 } | LC_ALL
=C
sort -u |
1054 git cat-file
${var_have_git_260:+--buffer} --batch-check"${var_have_git_185:+=%(objectname)}" |
1055 awk '!/missing/ {num++; print $1 " " "refs/" substr("zzzzzzzzzzzz", 1, length(num)) "/" num}'
1062 # Everything else is more efficient if we do this first
1063 # The "--prune" option is the default since v1.5.0 but it serves as "documentation" here
1064 git pack-refs
--all --prune
1065 [ -e packed-refs
] ||
>>packed-refs
# should never happen...
1067 # If we have a logs directory or a worktrees directory expire the ref logs now
1068 # Note that Git itself does not use either --rewrite or --updateref, so neither do we
1069 ! [ -d logs
] && ! [ -d worktrees
] ||
eval git reflog expire
--all "${quiet:+>/dev/null 2>&1}" ||
:
1072 ! [ -e .gc_failed
] ||
exit 1
1073 rm -f .gc_in_progress
# make sure
1074 touch .gc_in_progress
# it's truly fresh
1075 rm -f bundles
/* objects
/pack
/pack-
*.bndl
1076 # This is perhaps a bit aggressive in that if we're suffering from "Push Pack Redux"
1077 # and somehow we get run again immediately after the run where "Push Pack Redux" happened
1078 # and we have garbage collection forced, there's just the barest, almost negligible,
1079 # possibility that the "Push Pack Redux" ref updates _still_ have not happened and we
1080 # should not be removing _r .keep files. None of the normal Girocco processing can
1081 # cause this. The second run of this script would have to use the force gc option
1082 # for it to even be possible in the first place. What's much more likely is that
1083 # the initial run of this script was somehow interrupted in the middle before it
1084 # could get rid of the _r .keep file itself in which case it's better to get rid of
1085 # it now to avoid keeping something around that would perturb our nice and neat gc
1086 rm -f objects
/pack
/pack-
*_r.keep
1087 # We will add .keep files for _u packs if and when we run phase III
1088 # Otherwise they need to not have any .keep files during phases I and II
1089 rm -f objects
/pack
/pack-
*_u.keep
1091 # We need to make sure that any non-Girocco (barely tolerated) Git object creation
1092 # activity will be able to "freshen" the pack containing a pre-existing object
1093 # that's being written. This really should not be necessary as the pre-receive
1094 # hook should make sure this takes place for any incoming pushes.
1095 # However, do it here anyway just in case.
1096 make_packs_ugw objects
/pack
1098 # This is only effective with Git v2.3.5 and later and it will only matter when
1099 # we are using one of the "internal_rev_list" modes of pack-objects
1100 # (the combine-packs.sh script never uses any of those modes)
1101 # The "git repack" and "git prune" commands always sets this internally themselves
1102 # It makes no difference if there's no repository corruption
1103 GIT_REF_PARANOIA
=1 && export GIT_REF_PARANOIA
1105 # All of the options we might want to use with pack-objects were supported
1106 # at some point prior to Git version v1.6.6 which is the minimum version that
1107 # Girocco now requires. Except for one (--use-bitmap-index). Several of them
1108 # are "boiler plate" options we always want to use so we bundle them up here.
1109 pkopt
="--delta-base-offset --keep-true-parents --non-empty --all-progress-implied"
1110 # We want to use --include-tag, but before Git v2.10.1 it would leave out
1111 # "middle" tags (e.g. a tag of a tag of a commit would omit the tagged tag)
1112 # See http://repo.or.cz/git.git/b773ddea2cd3b08c for details
1113 # ("pack-objects: walk tag chains for --include-tag", 2016-09-07, v2.10.1)
1114 # This is not a free check as it matches all refs against refs/tags/ then
1115 # peels all the annotated tags and checks for inclusion. The situation in
1116 # which it would add a tag that was not already included by a reachability
1117 # trace that included tag starting points can only occur if a new tag gets
1118 # pushed during gc pointing to something that would have been packed anyway.
1119 # But, it could happen and, really, campared to gc as a whole it's not that
1120 # expensive to perform (provided we do not get an unconnected pack).
1121 [ -z "$var_have_git_2101" ] || pkopt
="$pkopt --include-tag"
1122 pkopt
="$pkopt ${quiet:---progress} $packopts"
1124 # The git pack-objects command only supports bitmaps if all objects are being
1125 # packed (the "--all" option) and the "--stdout" option is NOT being used.
1126 # Additionally, while packing, if any encountered reachable objects are
1127 # determined to be "not wanted" then no bitmap index will be written anyway.
1128 # While it is theoretically possible that a project with a non-empty alternates
1129 # file ends up packing all objects (because it does not actually use any of the
1130 # objects found in the alternates), it's very unlikely. And, in the unlikely
1131 # event that did occur, clients would see a message about only using one bitmap
1132 # because Git can only use one bitmap at a time and at least one of the
1133 # alternates is bound to have a bitmap. Therefore if we see a non-empty
1134 # alternates file, we disable writing bitmaps which avoids the warning and any
1135 # possibility of a client warning as well. Also if we are running anything
1136 # before Git v2.1.0 (the effective version for repack.writeBitmaps=true) then
1137 # we also always disable bitmap writing.
1139 [ -z "$var_have_git_210" ] || wbmopt
="--write-bitmap-index"
1140 # More recent versions of pack-objects have optimizations when not using the
1141 # --local option. If we do not have any alternates it's a pointless option.
1142 # If we do have alternates we need to skip writing a bitmap and we cannot
1143 # have a bundle since it must contain all objects.
1144 if [ -n "$isfork" ]; then
1158 [ -n "$wbmopt" ] || wbmstr
=" (bitmaps disabled)"
1159 progress
"~ [$proj] running primary full gc pack-objects$wbmstr"
1162 ! has_forks_with_alternates
"$proj" || gotforks
=1
1164 # To avoid "Push Pack Redux" (see README-GC), after collecting the initial
1165 # preexisting non-keep pack list, we rename them so that an incoming push
1166 # pack cannot possibly experience a pack name collision. Git does not require
1167 # use of the "default" pack names, simply that the proper extensions are used.
1168 # We rename to insert an "_r" just before the extension to avoid "Push Pack Redux"
1169 # name collisions. Later on we may create an "unreachable" pack for hard-linking
1170 # down into forks and it will have an "_u" inserted just before its extension.
1171 packlist
="$(list_packs -C objects/pack --exclude-no-idx --exclude-keep --quiet .)" ||
:
1173 for oldpack
in $packlist; do
1174 oldpack
="${oldpack#pack-}"
1175 oldpack
="${oldpack%.pack}"
1176 [ -f "objects/pack/pack-$oldpack.pack" ] ||
{
1177 echo >&2 "[$proj] unable to list old pack files"
1180 if [ "${oldpack#*[!0-9a-fA-F]}" != "$oldpack" ]; then
1181 # names not exclusively hexadecimal do not need renaming
1182 oldpacks
="${oldpacks:+$oldpacks }$oldpack"
1185 rename_pack
"objects/pack/pack-$oldpack" "objects/pack/pack-${oldpack}_r" ||
{
1186 echo >&2 "[$proj] unable to rename old pack files"
1189 # If the oldpack has a .keep now it means a "Push Pack Redux" is actually
1190 # in progress at this moment and we need to .keep the renamed pack,
1191 # otherwise no "Push Pack Redux" has started yet or it has already finished.
1192 # In either case we're okay because if it's just finished then all ref
1193 # changes have already been made so we don't need a .keep and we will
1194 # see the ref changes and grab all the objects via a reachability trace.
1195 # If it hasn't started yet that's okay because we're done moving that
1196 # name so a complete pack will appear under the old name that we'll
1198 if [ -f "objects/pack/pack-$oldpack.keep" ]; then
1199 echo "Push Pack Redux" >"objects/pack/pack-${oldpack}_r.keep"
1201 oldpacks
="${oldpacks:+$oldpacks }${oldpack}_r"
1205 # We wish to keep deltas from our last full pack so if we're not redeltaing
1206 # then make sure the .pack associated with the .bitmap has a newer mod time
1207 # (If there is no .bitmap then touch the pack with the most objects instead.)
1208 if [ -z "$newdeltas" ]; then
1209 bmpack
="$(list_packs --exclude-no-bitmap --exclude-no-idx --max-matches 1 objects/pack)"
1210 [ -n "$bmpack" ] || bmpack
="$(list_packs --exclude-no-idx --max-matches 1 --object-limit -1 --include-boundary objects/pack)"
1211 if [ -n "$bmpack" ] && [ -f "$bmpack" ] && [ -s "$bmpack" ]; then
1213 touch -c "$bmpack" 2>/dev
/null ||
:
1214 # We must touch .gc_in_progress here to avoid $bmpack looking
1215 # like it's been "freshened" when redundant packs are removed
1216 # It's okay if they have the same mod time, but POSIX does not
1217 # guarantee an ordering for the "touching" that occurs which is
1218 # why this must be a separate command but needs no "sleep 1"
1219 touch .gc_in_progress
1223 # Now we need to make sure that any "freshening" that takes place will actually
1224 # result in a "newer" modification time than the .gc_in_progress file now has
1227 # We run git pack-objects from the repack subdirectory so we can force
1228 # optimized packs to be generated even for repositories that do not have any
1230 packs
="$(git --git-dir=repack pack-objects </dev/null \
1231 $pkopt --all $newdeltas $lclopt ${wbmopt:---honor-pack-keep} "$@
" repack/alt/pack/pack)"
1233 [ $packcnt -eq 1 ] || makebndl
=
1239 progress
"~ [$proj] running supplementary gc pack-objects"
1241 # Add the "supplementary" refs
1242 compute_extra_reachables
>>repack
/packed-refs
1244 # Subtract the primary refs
1245 GIT_ALTERNATE_OBJECT_DIRECTORIES
="$PWD/repack/alt"
1246 export GIT_ALTERNATE_OBJECT_DIRECTORIES
1248 # For this one we MUST use --local and MUST NOT use --write-bitmap-index
1249 # However, if there is a "logs" subdirectory we need to use --reflog
1250 # We do add it, just in case, if the linked working trees dir is present
1251 # We do not add --indexed-objects as that requires v2.2.0 and it's unclear
1252 # if it properly includes linked working tree index files or not. The
1253 # above compute_extra_reachables has already included all index trees (thereby
1254 # providing proper --indexed-objects support for all Git versions) making the
1255 # option completely unnecessary.
1257 ! [ -d logs
] && ! [ -d worktrees
] || rflopt
=--reflog
1258 spacks
="$(git --git-dir=repack pack-objects </dev/null \
1259 $pkopt --honor-pack-keep --all $rflopt $newdeltas --local "$@
" repack/alt/pack/pack)"
1265 # There's nothing to do for Phase III unless we have forks that refer to our
1266 # project from their alternates file
1269 if [ -n "$gotforks" ]; then
1271 progress
"~ [$proj] running keep-unreachable gc pack-objects for forks"
1273 # If we are a fork, any pre-existing _u packs need to have a .keep
1274 # for this phase and be added to the hlpacks list otherwise (we are
1275 # not a fork) pre-existing _u packs are anomalies to be treated like
1276 # regular non-_u packs
1277 if [ -n "$isfork" ]; then
1278 for upack
in $
(find -L objects
/pack
-mindepth 1 -maxdepth 1 -name "pack-$octet20*_u.pack" -print); do
1279 upack
="${upack%.pack}"
1280 [ -e "$upack.keep" ] ||
echo "unreachable" >"$upack.keep"
1281 hlpacks
="${hlpacks:+$hlpacks }${upack#objects/pack/pack-}"
1284 # Using either --no-reuse-delta or --no-reuse-object together with the
1285 # --keep-unreachable option is a very, very, very bad idea when good
1286 # packs are the desired outcome. If newdeltas are being generated
1287 # then we pack to a temp name, and use combine-packs.sh to get a better
1288 # pack as the result to avoid making a bad --keep-unreachable pack
1290 [ -z "$newdeltas" ] || pfx
="ku"
1291 upacks
="$(git --git-dir=repack pack-objects </dev/null \
1292 $pkopt --honor-pack-keep --all $rflopt --keep-unreachable --local "$@
" repack/alt/pack/${pfx}pack)"
1293 if [ -n "$upacks" ] && [ -n "$newdeltas" ]; then
1294 progress
"~ [$proj] rebuilding keep-unreachable pack deltas"
1297 printf "repack
/alt
/pack
/${pfx}pack-
%s.pack
\n" $oldupacks |
1298 run_combine_packs --names --weak-naming --non-empty --all-progress-implied ${quiet:---progress} \
1299 $packopts $newdeltas "$@
" repack/alt/pack/pack)"
1300 eval rm -f "$(printf \""repack
/alt
/pack
/${pfx}pack-
%s.
*"\"" " $oldupacks)"
1302 for upack
in $upacks; do
1303 rename_pack
"repack/alt/pack/pack-$upack" "repack/alt/pack/pack-${upack}_u"
1305 rm -f objects
/pack
/pack-
*_u.keep
1306 [ -z "$hlpacks" ] && [ -z "$upacks" ] ||
1307 progress
"~ [$proj] hard-linking keep-unreachable pack(s) into immediate child forks"
1309 # We have to update the lastparentgc time in the child forks even if they do not get any
1310 # new "unreachable packs" because they need to run gc just in case the parent now has some
1311 # objects that used to only be in the child so they can be removed from the child.
1312 # For example, a "patch" might be developed first in a fork and then later accepted into
1313 # the parent in which case the objects making up the patch in the child fork are now
1314 # redundant (since they're now in the parent as well) and need to be removed from the
1315 # child fork which can only happen if the child fork runs gc.
1316 lastparentgc
="$(date "$datefmt")"
1318 # It is enough to copy objects just one level down and get_repo_list
1319 # takes a regular expression (which is automatically prefixed with '^')
1320 # so we can easily match forks exactly one level down from this project
1322 get_repo_list
"$forkdir/[^/:][^/:]*:" |
1324 # Ignore forks that do not exist or are symbolic links
1325 ! [ -L "$cfg_reporoot/$fork.git" ] && [ -d "$cfg_reporoot/$fork.git" ] ||
1327 # Or do not have a non-zero length alternates file
1328 [ -s "$cfg_reporoot/$fork.git/objects/info/alternates" ] ||
1331 # Match hlpacks in parent project if any
1332 if [ -n "$hlpacks" ]; then
1333 mkdir
-p "$cfg_reporoot/$fork.git/objects/pack"
1334 eval ln -f "$(printf '"objects
/pack
/pack-
%s.pack
" ' $hlpacks)" \
1335 "$(printf '"objects
/pack
/pack-
%s.idx
" ' $hlpacks)" \
1336 '"$cfg_reporoot/$fork.git/objects/pack/"'
1339 # Match upacks in repack/alt area if any
1340 if [ -n "$upacks" ]; then
1341 mkdir
-p "$cfg_reporoot/$fork.git/objects/pack"
1342 eval ln -f "$(printf '"repack
/alt
/pack
/pack-
%s_u.pack
" ' $upacks)" \
1343 "$(printf '"repack
/alt
/pack
/pack-
%s_u.idx
" ' $upacks)" \
1344 '"$cfg_reporoot/$fork.git/objects/pack/"'
1347 if ! [ -e "$cfg_reporoot/$fork.git/.needsgc" ]; then
1348 # Trigger a mini gc in the fork if it now has too many packs
1349 packs
="$(list_packs --quiet --count --exclude-no-idx --exclude-keep "$cfg_reporoot/$fork.git
/objects
/pack
")" ||
:
1350 if [ -n "$packs" ] && [ "$packs" -ge 20 ]; then
1351 >"$cfg_reporoot/$fork.git/.needsgc"
1354 [ -z "$runupdate" ] || git
--git-dir="$cfg_reporoot/$fork.git" update-server-info
1355 # Update the fork's lastparentgc date (must be more recent than $gcstart)
1356 git
--git-dir="$cfg_reporoot/$fork.git" config gitweb.lastparentgc
"$lastparentgc"
1360 # Now move any primary/supplementary packs back into objects/pack
1361 # then drop any "unfreshened" redundant packs and clear repack/alt
1363 # First make sure the primary pack(s) have the most recent mod time
1364 if [ -n "$packs" ]; then
1365 [ -z "$spacks" ] ||
sleep 1
1366 printf 'repack/alt/pack/pack-%s.pack\n' $packs |
xargs touch -c 2>/dev
/null ||
:
1369 # Move the packs into place
1370 for pack
in $packs $spacks; do
1371 rename_pack
"repack/alt/pack/pack-$pack" "objects/pack/pack-$pack"
1374 # It's possible that one of the $oldpacks had a .bitmap, got renamed (along
1375 # with its .bitmap) and then got "freshened" causing us to not remove it
1376 # However, if $wbmopt is set we most likely now have TWO .bitmap packs!
1377 # This can produce ugly warnings we don't want and possibly get the wrong
1378 # bitmap used since only one .bitmap file can ever be used by Git.
1379 # If this has happened, the .bitmap we want to discard will always have
1380 # an _r infix so we can just zap any such now since it will leave the pack.
1381 [ -z "$wbmopt" ] ||
rm -f objects
/pack
/pack-
*_r.bitmap ||
:
1383 # Remove the redundant packs that have not since been "freshened"
1384 # This does not completely eliminate the race condition window (Girocco's own
1385 # activites -- gc/fetch/receive are immune to the race) but it substantially
1386 # shrinks it down to just the time after the find but before the following rm
1388 [ -z "$oldpacks" ] ||
1389 printf 'objects/pack/pack-%s.pack\n' $oldpacks |
1390 LC_ALL
=C
sort >repack
/oldpacks
1391 find -L objects
/pack
-maxdepth 1 -type f
-name "pack-$octet20*.pack" -newer .gc_in_progress
-print |
1392 LC_ALL
=C
sort >repack
/freshened
1393 deadpacks
="$(LC_ALL=C join -v 1 repack/oldpacks repack/freshened | LC_ALL=C sed 's/\.pack$//')"
1394 [ -z "$deadpacks" ] ||
1395 eval echo "$(printf '"%s
".* ' $deadpacks)" |
xargs rm -f ||
:
1397 # No need for this anymore
1398 rm -rf repack
/alt objects
/pack
/repack
1399 unset GIT_ALTERNATE_OBJECT_DIRECTORIES
1405 progress
"~ [$proj] running gc prune-packed"
1407 # We do not want the redundant packs or any new "--keep-unreachable" pack(s) to be
1408 # present while running prune-packed. We try to guarantee that any loose object
1409 # that's unreachable persists for at least one $Girocco::Config::min_gc_interval
1410 # (not withstanding administrator interference to force earlier gc to occur).
1411 # If we were to include the redundant/keep-unreachable pack(s) when running
1412 # prune-packed and a loose unreachable object happened to be duplicated in one
1413 # of them we would end up removing it too soon and void our guarantee.
1414 git prune-packed
$quiet
1416 progress
"~ [$proj] running loose objects gc pack-objects"
1418 # Although Git v2.10.0 and later support a --pack-loose-unreachable option,
1419 # we MUST NOT use it for these reasons:
1420 # 1) We're not interested in expensive "unreachable" at this point, only "loose"
1421 # 2) It produces simply horrid packs about 3.8x times larger than they should be
1422 # 3) We don't require anything more than Git v1.6.6
1423 lpacks
="$(run_combine_packs </dev/null --names --loose --weak-naming --non-empty --all-progress-implied ${quiet:---progress} $packopts $newdeltas "$@
")"
1425 if [ -n "$lpacks" ]; then
1426 # Make sure any primary pack(s) have a more recent mod time than "unreachable" objects packs
1427 if [ -n "$packs" ]; then
1429 printf 'objects/pack/pack-%s.pack\n' $packs |
xargs touch -c 2>/dev
/null ||
:
1431 # We need to identify these packs later so we don't combine_packs them
1432 for objpack
in $lpacks; do
1433 rename_pack
"objects/pack/pack-$objpack" "objects/pack/pack-${objpack}_o" ||
:
1435 # Finally zap the corresponding loose objects
1436 progress
"~ [$proj] running packed loose objects gc prune-packed"
1437 git prune-packed
$quiet
1440 ! [ -e .gc_failed
] ||
exit 1
1441 # These, if they exist, are now meaningless and need to be removed
1442 rm -f gfi-packs .needsgc .svnpack .svnpackgc
1444 # Make sure this stays up to date
1445 git update-server-info
1447 # We must make loose objects group writable so that they
1448 # can be freshened by other pushers. Technically we need only do this for
1449 # push projects but to enable mirror projects to be more easily converted to
1450 # push projects, we go ahead and do it for all projects.
1451 # By the time we get here we really shouldn't have any of these, but just in case.
1452 { find -L objects
/$octet -type f
-name "$octet19*" -exec chmod ug
+w
'{}' + ||
:; } 2>/dev
/null
1454 # darcs:// mirrors have a xxx.log file that will grow endlessly
1455 # if this is a mirror and the file exists, shorten it to 10000 lines
1456 # also take this opportunity to optimize the darcs repo
1457 if ! [ -e .nofetch
] && [ -n "$cfg_mirror" ]; then
1458 url
="$(config_get baseurl)" ||
:
1459 case "$url" in darcs
://*)
1460 if [ -n "$cfg_mirror_darcs" ]; then
1462 basedarcs
="$(basename "${url#darcs:/}")"
1463 if [ -f "$basedarcs.log" ]; then
1464 tail -n 10000 "$basedarcs.log" >"$basedarcs.log.$$"
1465 mv -f "$basedarcs.log.$$" "$basedarcs.log"
1467 if [ -d "$basedarcs.darcs" ]; then
1469 cd "$basedarcs.darcs"
1470 # without show_progress suppress non-error output
1471 [ -n "$show_progress" ] ||
exec >/dev
/null
1472 # Note that this does not optimize _darcs/inventories/ :(
1480 # Create a matching .bndl header file for the all-in-one pack we just created
1481 # but only if we're not a fork (otherwise the bundle would not be complete)
1482 # and we are running at least Git version 1.7.2 (pack_is_complete always fails otherwise)
1483 if [ -n "$makebndl" ] && [ -n "$var_have_git_172" ]; then
1484 # There should only be one pack in $packs but do some checking...
1485 # The one we just created will have a .idx and will NOT have a .keep
1486 progress
"~ [$proj] creating downloadble bundle header"
1489 IFS
= read -r curhead
<repack
/HEAD.orig ||
:
1491 [ -s "objects/pack/pack-$packs.pack" ] &&
1492 [ -s "objects/pack/pack-$packs.idx" ] &&
1493 ! [ -e "objects/pack/pack-$packs.keep" ] &&
1494 pkhead
="$(pack_is_complete "$PWD/objects
/pack
/pack-
$packs.pack
" \
1495 "$PWD/repack
/packed-refs.orig
" "$curhead")"
1497 pkbase
="objects/pack/pack-$packs"
1499 if [ -n "$pkbase" ] && [ -n "$pkhead" ]; then
1502 case "$curhead" in "ref: refs/"?
*|
"ref:refs/"?
*|
"refs/"?
*)
1503 symref
="${curhead#ref:}"
1504 symref
="${symref# }"
1507 [ -z "$cfg_httpbundleurl" ] || bndlurl
=" url=$cfg_httpbundleurl/$proj.git/clone.bundle"
1508 echo "# v2 git bundle"
1509 LC_ALL
=C
sed -ne "/^$octet20$hexdig* refs\/[^ $tab]*\$/ p" <repack
/packed-refs.orig
1510 if [ -n "$symref" ]; then
1511 printf "$pkhead HEAD\0symref=HEAD:%s%s\n" "$symref" "$bndlurl"
1513 if [ -n "$bndlurl" ]; then
1514 printf "$pkhead HEAD\0%s\n" "${bndlurl:# }"
1521 bndletag
="$("$cfg_basedir/bin
/rangecgi
" --etag -m 1 "$pkbase.bndl
" "$pkbase.pack
")" ||
:
1522 bndlsha
="$(printf '%s' "$bndletag" | git hash-object --stdin)" ||
:
1523 if [ -n "$bndletag" ]; then
1524 case "$bndlsha" in $octet20*)
1525 bndlshatrailer
="${bndlsha#????????}"
1526 bndlshaprefix
="${bndlsha%$bndlshatrailer}"
1527 bndlname
="$(TZ=UTC date +%Y%m%d_%H%M%S)-${bndlshaprefix:-0}"
1528 [ -d bundles
] || mkdir bundles
1529 echo "${pkbase#objects/pack/}.bndl" >"bundles/$bndlname"
1530 echo "${pkbase#objects/pack/}.pack" >>"bundles/$bndlname"
1531 ln -s -f -n "$bndlname" bundles
/latest
1537 # Record the size of this repo as the sum of its clone packed-refs + *.pack sizes as 1024-byte blocks
1538 eval "reposizek=$(( $(
1539 echo 0 $(du -k repack/packed-refs.orig $(printf 'objects/pack/pack-%s.pack ' $packs) 2>/dev/null |
1540 LC_ALL=C awk '{print $1}') |
1541 LC_ALL=C sed -e 's/ / + /g') ))"
1542 config_set_raw girocco.reposizek
"${reposizek:-0}"
1544 # Now we're finally done with this
1547 # We didn't used to do anything about rerere or worktrees but we're
1548 # trying to make nice with linked working trees these days :)
1549 # Maybe even non-bare repositories too, but *shush* about those ;)
1550 if [ -n "$var_have_git_250" ] && [ -d worktrees
]; then
1551 # The value "3.months.ago" is hard-coded into gc.c rather than
1552 # having the default be in worktree.c so we must provide it if
1553 # we get nothing out of the gc.worktreePruneExpire config item
1554 # Prior to Git v2.6.0 the config item was gc.pruneworktreesexpire
1555 # however we just always use the newer name no matter what Git version
1556 expiry
="$(git config --get gc.worktreePruneExpire 2>/dev/null)" ||
:
1557 eval git worktree prune
--expire '"${expiry:-3.months.ago}"' "${quiet:+>/dev/null 2>&1}" ||
:
1559 # git rerere does it right and handles its own default/config'd expiration values
1560 ! [ -d rr-cache
] ||
eval git rerere gc
"${quiet:+>/dev/null 2>&1}" ||
:
1562 # We use $gcstart here to avoid a race where a push occurs during the gc itself
1563 # and the next future gc could be incorrectly skipped if we used the current
1564 # timestamp here instead
1565 config_set lastgc
"$gcstart"
1568 progress
"- [$proj] garbage check ($(date))"