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
17 packopts
="--depth=50 --window=50 --window-memory=${var_window_memory:-1g}"
18 quiet
=; [ -n "$show_progress" ] || quiet
=-q
21 [ "$cfg_permission_control" != "Hooks" ] ||
umask 000
25 if _result
="$(kill -0 "$1" 2>&1)"; then
26 # process exists and we have permission to signal it
29 case "$_result" in *"not permitted"*)
30 # we do not have permission to signal the process
33 # process does not exist
38 # A .lock file should only exist for much less than a second.
39 # If we see a stale lock file (> 1h old), remove it and then,
40 # just in case, wait 30 seconds for any process whose .lock
41 # we might have just removed (it's racy) to finish doing what
42 # should take much less than a second to do.
43 _stalelock
="$(find -L "$1.lock
" -maxdepth 1 -mmin +60 -print 2>/dev/null)" ||
:
44 if [ -n "$_stalelock" ]; then
49 if (set -C; >"$1.lock") 2>/dev
/null
; then
54 [ "$_try" != "p" ] ||
sleep 1
56 # cannot create lock file
60 # The pre-receive script creates one ref log file per push but we want them to
61 # be coalesced into one ref log file per day. We are guaranteed that any files
62 # we find to coalesce are NOT currently being written to since they are always
63 # written first as temporary files and then moved into place. We attempt to
64 # transfer the most recent modification time to the coalesced log file which
65 # would step on its mod time if it were being written to directly, but if we
66 # find per-process ref log files then it must be a push project and the only
67 # thing that would write directly to the main per-day log file would be a
68 # mirror project so there's actually no conflict.
69 # Also, if the clock is wonky (or was futzed with) we may have both YYYYMMDD
70 # and YYYYMMDD.gz present in which case combine them into YYYYMMDD
72 [ -d reflogs
] ||
return 0
74 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 |
75 while read -r rname
; do
76 if [ -e "$rname.gz" ]; then
77 if [ -s "$rname" ]; then
78 # Presumably the .gz file must have been created before the non-gz
79 # file since it had to be uncompressed at some point therefore
80 # we need to append the non-gz contents to it but keep the non-gz
81 # contents timestamp so we rename to YYYYMMDD_ which will sort first
82 # and be picked up in the next step if we are interrupted in the middle.
83 # If a YYYYMMDD_ file already exists we append to it and transfer the
84 # timestamp. Finally we transfer the YYYYMMDD_ timestamp to the result
85 # and remove the YYYYMMDD_ temporary file leaving the result uncompressed.
86 if [ -e "${rname}_" ]; then
87 cat "$rname" >>"${rname}_"
88 touch -r "$rname" "${rname}_"
92 mv "$rname" "${rname}_"
94 gzip -d "$rname.gz" </dev
/null
95 [ -e "$rname" ] && ! [ -e "$rname.gz" ]
96 cat "${rname}_" >>"$rname"
97 touch -r "${rname}_" "$rname"
100 # Just remove the empty file to resolve the problem
105 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 |
106 while read -r rname
; do
107 logname
="${rname%%_*}"
108 # If someone's been futzing with the date, the file we want to
109 # append to could already have been compressed, so we just uncompress
110 # it here. The previous block guarantees we do not have both a compressed
111 # and uncompressed version present at the same time.
112 if [ -e "$logname.gz" ]; then
113 gzip -d "$logname.gz" </dev
/null
114 [ -e "$logname" ] && ! [ -e "$logname.gz" ]
116 cat "$rname" >>"$logname"
117 touch -r "$rname" "$logname"
119 if [ -e "$rname" ]; then
121 echo "! [$proj] failed to remove $rname" >&2
122 exit 1 # will only exit subshell created by "|"
128 # Remove any files in reflogs that are older than $cfg_reflogs_lifetime days
130 [ -d reflogs
] ||
return 0
131 exp
="$(( ${cfg_reflogs_lifetime:-1} * 1440 ))"
132 [ $exp -gt 0 ] || exp
=1440
133 [ $exp -le 43200 ] || exp
=43200
134 find -L reflogs
-maxdepth 1 -type f
-mmin "+$exp" -exec rm -f '{}' + ||
:
137 # Compact any reflogs that are not today's UTC date unless a .gz version exists
139 [ -d reflogs
] ||
return 0
140 _td
="reflogs/$(TZ=UTC date '+%Y%m%d')"
141 find -L reflogs
-maxdepth 1 -type f
-name "[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]" -print |
142 while read -r rname
; do
143 [ "$rname" != "$_td" ] ||
continue
144 ! [ -e "$rname.gz" ] ||
continue
145 gzip -9 "$rname" </dev
/null
149 # return true if there's more than one objects/pack-<sha>.pack file or
150 # ANY sha-1 files in objects
152 _packs
=$
(find -L objects
/pack
-type f
-name "pack-$octet20*.pack" -print |
head -n 2 | LC_ALL
=C
wc -l)
153 if [ $_packs != 1 ] && [ $_packs != 0 ]; then
156 _objs
=$
(find -L objects
/$octet -type f
-name "$octet19*" -print 2>/dev
/null |
head -n 1 | LC_ALL
=C
wc -l)
160 # make sure combine-packs uses the correct Git executable
161 run_combine_packs
() {
162 PATH
="$var_git_exec_path:$cfg_basedir/bin:$PATH" @basedir@
/jobd
/combine-packs.sh
"$@"
165 # combine the input pack(s) into a new pack (or possibly packs if packSizeLimit set)
166 # input pack names are read from standard input one per line delimited by the first
167 # ':', ' ' or '\n' character on the line (which allows gfi-packs to be read directly)
168 # all arguments, if any, are passed to pack-objects as additional options
169 # returns non-zero on failure AND creates .gc_failed in that case
172 find -L objects
/pack
-maxdepth 1 -type f
-name '*.zap*' -exec rm -f '{}' + ||
:
173 run_combine_packs
--replace "$@" $packopts --all-progress-implied $quiet --non-empty ||
{
180 # if the current directory is_gfi_mirror then repack all packs listed in gfi-packs
182 [ -n "$gfi_mirror" ] ||
return 0
183 [ -d objects
/pack
] ||
{ rm -f gfi-packs
; return 0; }
184 progress
"~ [$proj] redeltifying poor quality git fast-import packs"
185 combine_packs
--ignore-missing --no-reuse-delta <gfi-packs
190 # combine small packs into larger pack(s)
191 # we avoid any keep, bndl or bitmap packs
192 # if the optional argument is non-empty even a single small pack will be redeltad
193 combine_small_packs
() {
196 if [ -n "$1" ] && [ -n "$noreusedeltaopt" ]; then
199 _lpo
="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl"
200 _lpo
="$_lpo --exclude-sfx _u --exclude-sfx _o"
201 _lpo
="$_lpo --quiet --object-limit $var_redelta_threshold objects/pack"
203 _cnt
="$(list_packs --count $_lpo)" ||
:
204 test "${_cnt:-0}" -ge $_minsmallpacks
206 [ -n "$_didprogress" ] ||
{
207 progress
"~ [$proj] combining small packs into a single larger pack"
210 _newp
="$(list_packs $_lpo | combine_packs --names $noreusedeltaopt)"
211 _newc
="$(( $(echo "$_newp" | LC_ALL=C wc -w) ))"
212 # be paranoid and exit the loop if we haven't reduced the number of packs
213 [ $_newc -lt $_cnt ] ||
break
219 # Unfortunately git-svn lacks the ability to store newly fetched revisions as a pack.
220 # However, the fetch code conveniently sets .svnpack just before it runs git-svn fetch
221 # so that it's easy to find all the objects that have been fetched by git-svn and
222 # combine them into a pack. The --no-reuse-delta option is meaningless here since
223 # everything to be packed is a loose object and therefore not a delta so deltification
224 # will always take place.
226 [ -f .svnpack
] && [ -n "$svn_mirror" ] ||
return 0
228 mv -f .svnpack .svnpackgc
229 progress
"~ [$proj] combining loose git-svn objects into a pack"
230 _newp
="$(find -L objects/$octet -maxdepth 1 -type f -newer .svnpackgc -name "$octet19*" -print 2>/dev/null |
231 LC_ALL=C awk -F / '{print $2 $3}' |
232 run_combine_packs --objects --names $packopts --incremental --all-progress-implied $quiet --non-empty)" ||
{
233 mv -f .svnpackgc .svnpack
237 if [ -n "$_newp" ]; then
238 # remove the now-redundant loose objects -- this is always safe
239 # even during a concurrent push because a reprepare_packed_git
240 # will be triggered if an object that should be there is not
241 # found thereby finding it in the new pack instead
242 git prune-packed
$quiet
247 # HEADSHA="$(pack_is_complete /full/path/to/some.pack /full/path/to/packed-refs "$(cat HEAD)")"
249 # Must have a matching .idx file and a non-empty packed-refs file
250 [ -s "${1%.pack}.idx" ] ||
return 1
251 [ -s "$2" ] ||
return 1
257 "ref: refs/"?
*|
"ref:refs/"?
*|
"refs/"?
*)
258 _headmatch
="${3#ref:}"
259 _headmatch
="${_headmatch# }"
260 _headmatchpat
="$(echo "$_headmatch" | LC_ALL=C sed -e 's/\([.$]\)/\\\1/g')"
261 _headsha
="$(LC_ALL=C grep -e "^
$octet20$hexdig* $_headmatchpat\$
" <"$2" |
262 LC_ALL=C cut -d ' ' -f 1)"
263 case "$_headsha" in $octet20*) :;; *)
271 rm -rf pack_is_complete_test
272 mkdir pack_is_complete_test
273 mkdir pack_is_complete_test
/refs
274 mkdir pack_is_complete_test
/objects
275 mkdir pack_is_complete_test
/objects
/pack
276 echo "$_headsha" >pack_is_complete_test
/HEAD
277 ln -s "$1" pack_is_complete_test
/objects
/pack
/
278 ln -s "${1%.pack}.idx" pack_is_complete_test
/objects
/pack
/
279 ln -s "$2" pack_is_complete_test
/packed-refs
280 _count
="$(git --git-dir=pack_is_complete_test rev-list --count --all 2>/dev/null)" ||
:
281 rm -rf pack_is_complete_test
282 [ -n "$_count" ] ||
return 1
283 [ "$_count" -gt 0 ] 2>/dev
/null ||
return 1
287 # On return a "$lockf" will have been created that must be removed when gc is done
289 # be compatibile with gc.pid file from newer Git releases
293 if [ "$(createlock "$lockf")" ]; then
295 # 1) less than 12 hours old
296 # 2) contains two fields (pid hostname) NO trailing NL
297 # 3) the hostname is different OR the pid is still alive
298 # then we exit as another active process is holding the lock
299 if [ "$(find -L "$lockf" -maxdepth 1 -mmin -720 -print 2>/dev/null)" ]; then
302 read -r apid ahost ajunk
<"$lockf" ||
:
303 if [ "$apid" ] && [ "$ahost" ]; then
304 if [ "$ahost" != "$hn" ] || pidactive
"$apid"; then
310 echo >&2 "[$proj] unable to create gc.pid.lock file"
313 if [ -n "$active" ]; then
315 echo >&2 "[$proj] gc already running on machine '$ahost' pid '$apid'"
318 printf "%s %s" "$$" "$hn" >"$lockf.lock"
319 chmod 0664 "$lockf.lock"
320 mv -f "$lockf.lock" "$lockf"
323 # Create a repack subdirectory such that running repack in it will pack the
324 # same things that a pack in the normal directory would except that the pack
325 # is guaranteed to be generated in an optimized order by adding a suitable
326 # synthesized ref in the refs/tags namespace (yes, pack-objects.c really does
327 # behave differently depending on the contents of the refs/tags namespace).
328 # Before calling this, pack-refs --all MUST be performed or the wrong pack
329 # will end up being made.
331 # If a ref deletion is pushed after making the repack subdir but before the
332 # the actual repack, the discarded objects will be packed -- no big deal,
333 # they'll get discarded the next time gc runs.
335 # If a fast-forward ref update is pushed after making the repack subdir but
336 # before the actual repack, it will be picked up and the new objects packed
337 # (subject to the normal git repack race about picking such updates up).
339 # If a non-fast-forward ref update is pushed after making the repack subdir but
340 # before the actual repack, it will be picked up like a fast-forward update but
341 # the discarded objects will be included like a ref deletion (until the next
342 # scheduled gc takes place).
344 # We retain a copy of the original packed-refs file as repack/packed-refs.orig
345 # If ref deletions come in while we're repacking, the original packed-refs
346 # file will be modified, but we'll still pack the deleted ref(s).
347 # If the packed-refs.orig file is used to create the bundle header we avoid
348 # a situation where the bundle contains a ref state that never actually
349 # existed in reality (for example a new branch is pushed and then an old
350 # branch deleted afterwards -- the deletion would show up in the bundle
351 # because it will cause the original packed-refs file to be re-written, but
352 # the new branch creation will not unless we do another pack-refs which might
353 # lead to having in incomplete bundle). Therefore we want to keep a copy of
354 # the original packed-refs file around. We do the same thing for HEAD.
356 # It's possible that the "objects" subdirectory is a symbolic link.
357 # Git does support this. However, during the repacking process, new packs
358 # will be created in repack/alt/pack and then moved into objects/pack.
359 # In order for this to work seemlessly, they must both be on the same
360 # filesystem. But when objects (or even objects/pack) is a symbolic link they
361 # might not be. For this reason a "repack" subdirectory is created under
362 # objects/pack and the repack/alt/pack directory symbolicly linked to it.
364 # Git allows not just HEAD to be a symbolic-ref, but any ref anywhere in the
365 # refs namespace. We are concerned about ref name collisions and getting the
366 # right tag set to get an optimal pack. We can safely duplicate the ref space
367 # under refs/heads, refs/notes and refs/remotes without any risk of unwanted
368 # collisions and this will likely make over 99%+ of all symbolic refs found
369 # in the wild work properly. Girocco itself never creates any symbolic refs
370 # inside the refs namespace; this is a nod to simultaneously using a Girocco
371 # repository for other purposes.
373 ! [ -d repack
] ||
rm -rf repack
374 ! [ -d repack
] ||
{ echo >&2 "[$proj] cannot remove repack subdirectory"; exit 1; }
375 [ -d objects
/pack
] || mkdir
-p objects
/pack
376 ! [ -d objects
/pack
/repack
] ||
rm -rf objects
/pack
/repack
377 ! [ -d objects
/pack
/repack
] ||
{ echo >&2 "[$proj] cannot remove objects/pack/repack subdirectory"; exit 1; }
378 mkdir repack repack
/refs repack
/alt objects
/pack
/repack
379 [ -d info
] || mkdir info
380 ln -s ..
/config repack
/config
381 ln -s ..
/info repack
/info
382 ln -s ..
/objects repack
/objects
383 ln -s "$PWD/objects/pack/repack" repack
/alt
/pack
384 ln -s ..
/..
/refs repack
/refs
/refs
385 _lines
=$
(( $
(LC_ALL
=C
wc -l <packed-refs
) ))
386 cat HEAD
>repack
/HEAD.orig
387 cat packed-refs
>repack
/packed-refs.orig
388 if [ $
(LC_ALL
=C
wc -l <repack
/packed-refs.orig
) -ne "$_lines" ]; then
389 echo >&2 "[$proj] error: make_repack_dir failed original packed-refs line count sanity check"
392 # Note: Git v1.5.0 introduced the "# pack-refs with:" header line for the packed-refs file
393 sed '/^# pack-refs/d; s, refs/, refs/!/,' <repack
/packed-refs.orig
>repack
/packed-refs
394 headref
="$(git rev-parse --verify --quiet HEAD)" ||
:
395 if [ -n "$headref" ]; then
396 echo "$headref refs/!=/HEAD" >>repack
/packed-refs
397 echo "$headref refs/heads/!" >>repack
/packed-refs
398 _lines
=$
(( $_lines + 2 ))
400 if [ $
(( $
(LC_ALL
=C
wc -l <repack
/packed-refs
) + 1 )) -ne "$_lines" ]; then
401 echo >&2 "[$proj] error: make_repack_dir failed packed-refs initial line count sanity check"
404 sed -n '\, refs/heads/,p; \, refs/notes/,p; \, refs/remotes/,p' <repack
/packed-refs.orig
>>repack
/packed-refs
405 _newlines
="$(( $(LC_ALL=C wc -l <repack/packed-refs) ))"
406 if [ $
(( $_newlines + 1 )) -lt "$_lines" ]; then
407 echo >&2 "[$proj] error: make_repack_dir failed packed-refs extra line count sanity check"
411 optref
="$(git rev-list -n 1 --all 2>/dev/null)" ||
:
412 if [ -n "$optref" ]; then
413 echo "$optref refs/tags/!" >>repack
/packed-refs
414 _lines
=$
(( $_lines + 1 ))
415 echo "$optref" >repack
/HEAD
417 cat HEAD
>repack
/HEAD
419 if [ $
(LC_ALL
=C
wc -l <repack
/packed-refs
) -ne "$_lines" ]; then
420 echo >&2 "[$proj] error: make_repack_dir failed packed-refs line count sanity check"
425 # Remove any crud that's been left behind by interrupted operations
426 # that did not clean up after themselves
428 # Remove any existing FETCH_HEAD
429 # There can only be a FETCH_HEAD if we've been fetching, not if we've been
430 # receiving pushes (those never create a FETCH_HEAD).
431 # And if we're fetching because we're a mirror, we know we're not fetching right
432 # now since jobd.pl never runs a project's fetch simultaneously with its gc.
433 # Therefore any existing FETCH_HEAD is junk. And it may be many megabytes if
434 # there were a lot of refs.
437 # remove any existing pack_is_complete_test or repack subdirectories
438 # If either exists when this function is called it's crud
439 rm -rf pack_is_complete_test repack objects
/pack
/repack
441 # Remove any stale pack remnants that are more than an hour old.
442 # Stale pack fragments are defined as any pack-<sha1>.ext where .ext is NOT
443 # .pack AND the corresponding .pack DOES NOT exist. A bunch of stale
444 # pack-<sha1>.idx files without their corresponding .pack files are worthless
445 # and just waste space. Normally there shouldn't be any remnants but actually
446 # this can happen when things are interrupted at just the wrong time.
447 # Note that the objects/pack directory is created by git init and should
449 find -L objects
/pack
-maxdepth 1 -type f
-mmin +60 -name "pack-$octet20*.?*" -print |
450 LC_ALL
=C
sed -e 's/^objects\/pack\/pack-//; s/\..*$//' | LC_ALL
=C
sort -u |
451 while read packsha
; do
452 ! [ -e "objects/pack/pack-$packsha.pack" ] ||
continue
453 rm -f "objects/pack/pack-$packsha".?
*
456 # Remove any stale tmp reflogs files that are more than one hour old.
457 # Since they are created only while the pre-receive hook is running and
458 # all it does is process a bunch of refs passed to it on standard input
459 # it's inconceivable that it would ever take as much as an hour to run.
460 if [ -d reflogs
]; then
461 find -L reflogs
-maxdepth 1 -type f
-mmin +60 -name "tmp_*" -exec rm -f '{}' + ||
:
464 # Remove any stale object tmp_obj_* files that are more than 3 hours old.
465 # Really these files should only exist very briefly so there shouldn't be any
466 # but things happen that can end up leaving them behind.
467 find -L objects
/$octet -maxdepth 1 -type f
-mmin +180 -name "tmp_obj_?*" -exec rm -f '{}' + 2>/dev
/null ||
:
469 # Remove any stale pack .keep files that are more than 12 hours old.
470 # We don't do anything to create any permanent pack .keep files, so they must
471 # be remnants from some failed push or something. Removing the .keep will
472 # allow the pack to be properly repacked.
473 find -L objects
/pack
-maxdepth 1 -type f
-mmin +720 -name "pack-$octet20*.keep" -exec rm -f '{}' + ||
:
475 # Remove any stale tmp_pack_*, tmp_idx_*, tmp_bitmap_*, packtmp-* or .tmp-*-pack* files
476 # that are more than 12 hours old.
477 find -L objects
/pack
-maxdepth 1 -type f
-mmin +720 \
( \
478 -name "tmp_pack_?*" -o -name "tmp_idx_?*" -o -name "tmp_bitmap_?*" -o \
479 -name "packtmp-?*" -o -name ".tmp-?*-pack*" \
480 \
) -exec rm -f '{}' + ||
:
482 # Remove any stale incoming-* object quarantine directories that are
483 # more than 12 hours old. These are new with Git >= 2.11.0.
484 find -L objects
-maxdepth 1 -type d
-name 'incoming-?*' -mmin +720 \
485 -exec rm -rf '{}' + ||
:
487 # Remove any stale shallow_* files that are more than 12 hours old.
488 # These can be left behind by Git >= 1.8.4.2 and < 2.0.0 when a client
489 # requests a shallow clone.
490 find -L .
-maxdepth 1 -type f
-mmin +720 -name "shallow_?*" -exec rm -f '{}' + ||
:
492 # Remove any stale *.temp files in the objects area that are more than 12 hours old.
493 # This can be stale sha1.temp, or stale *.pack.temp so we kill all stale *.temp.
494 find -L objects
-type f
-mmin +720 -name "*.temp" -exec rm -f '{}' + ||
:
496 # Remove any stale *.lock files in the htmlcache area that might have been left
497 # behind after an abnormal exit during an attempt to update a cached file and
498 # are more than 1 hour old.
499 ! [ -d htmlcache
] ||
find -L htmlcache
-type f
-mmin +60 -name "*.lock" -exec rm -f '{}' + ||
:
501 # Remove any stale git-svn temp files that are more than 12 hours old.
502 # The git-svn process creates temp files with random 10 character names
503 # in the root of $GIT_DIR. Unfortunately they do not have a recognizable
504 # prefix, so we just have to kill any files with a 10-character name. We
505 # do this only for git-svn mirrors. All characters are chosen from
506 # [A-Za-z0-9_] so we can at least check that and fortunately the only
507 # collision is 'FETCH_HEAD' but that shouldn't matter.
508 # There may also be temp files with a Git_ prefix as well.
509 if [ -n "$svn_mirror" ]; then
510 _randchar
='[A-Za-z0-9_]'
511 _randchar2
="$_randchar$_randchar"
512 _randchar4
="$_randchar2$_randchar2"
513 _randchar10
="$_randchar4$_randchar4$_randchar2"
514 find -L .
-maxdepth 1 -type f
-mmin +720 -name "$_randchar10" -exec rm -f '{}' + ||
:
515 find -L .
-maxdepth 1 -type f
-mmin +720 -name "Git_*" -exec rm -f '{}' + ||
:
518 # Remove any stale fast_import_crash_<pid> files that are more than 3 days old.
519 if [ -n "$gfi_mirror" ]; then
520 find -L .
-maxdepth 1 -type f
-mmin +4320 -name "fast_import_crash_?*" -exec rm -f '{}' + ||
:
523 # Remove any stale core or *.core or core.* files that are more than 3 days old.
524 find -L .
-maxdepth 1 -type f
-mmin +4320 \
( -name "core" -o -name "*.core" -o -name "core.*" \
) \
525 -exec rm -f '{}' + ||
:
529 ## Garbage Collection Types
531 ## There are two kinds of possible garbage collection (gc) operations:
533 ## 1. A normal, full gc
536 ## If the full garbage collection interval has expired (or gc has never been
537 ## run), then a normal, full gc will take place. Otherwise, a "mini" gc will
538 ## take place if the file .needsgc exists.
540 ## A "mini" gc is similar to "git gc --auto" in that it may not end up actually
541 ## doing anything unless the right conditions are present so it's not a burden
542 ## to run it often. If the file .needsgc exists, a "mini" gc will occur at
543 ## the next opportunity.
545 ## Note, however, that the .nogc file suppresses ALL gc activity (normal or mini).
550 cd "$cfg_reporoot/$proj.git"
551 [ -d objects
/pack
] ||
{ rm -f gfi-packs
; mkdir
-p objects
/pack
; }
552 mirror_url
="$(get_mirror_url)" ||
:
554 ! is_svn_mirror_url
"$mirror_url" || svn_mirror
=1
556 if [ -f gfi-packs
] && [ -s gfi-packs
] && is_gfi_mirror_url
"$mirror_url"; then
560 # If git config --bool --get girocco.redelta is explicitly false then automatic
561 # redelta when there are less than $var_redelta_threshold objects will be suppressed.
562 # On the other hand, if git config --get girocco.redelta is "always" then, on a full
563 # gc only, for the final repack, deltas will always be recomputed.
564 # This can be set on a per-project basis to avoid unusual pathological gc behavior.
565 # Setting this will hurt efficiency of the affected repository.
566 # Note that fast-import packs ALWAYS get new deltas regardless of this setting.
567 noreusedeltaopt
="--no-reuse-delta"
568 [ "$(git config --bool --get girocco.redelta 2>/dev/null || :)" != "false" ] || noreusedeltaopt
=
570 [ "$(git config --get girocco.redelta 2>/dev/null || :)" != "always" ] || alwaysredelta
=1
572 # Extract any -f or -F or --no-reuse-object or --no-reuse-delta options
573 # to be compatible with the old and new gc.sh versions and avoid ugly argument
574 # duplication in process lists at the same time
575 # Any options found will override the "girocco.redelta" setting
578 while [ $idx -gt 0 ]; do
587 -F|
--no-reuse-object)
595 printf >&2 '%s\n' "bad non-option argument: $opt"
596 echo >&2 "(Did you perhaps intend to use a --xxx=yyy form?)"
599 [ -z "$opt" ] ||
set -- "$@" "$opt"
601 if [ -n "$alwaysredelta" ]; then
602 noreusedeltaopt
="--no-reuse-delta"
603 [ -z "$recompress" ] || noreusedeltaopt
="--no-reuse-object"
606 trap 'e=$?; rm -f .gc_in_progress; if [ $e != 0 ]; then echo "gc failed dir: $PWD" >&2; fi' EXIT
610 # date -R is linux-only, POSIX equivalent is '+%a, %d %b %Y %T %z'
611 datefmt
='+%a, %d %b %Y %T %z'
614 if [ "${force_gc:-0}" = "0" ] && check_interval lastgc
$cfg_min_gc_interval; then
615 if [ -e .needsgc
]; then
618 progress
"= [$proj] garbage check skip (last at $(config_get lastgc))"
622 if [ -e .nogc
]; then
623 progress
"x [$proj] garbage check disabled"
626 if [ -z "$isminigc" ] && [ -e .delaygc
] && [ -e .needsgc
]; then
627 # Eligible for a full gc but .delaygc is set so it would be skipped
628 # However .needsgc is also set so transform it into a mini instead
630 progress
"~ [$proj] garbage check delayed but checking mini because .needsgc"
633 if [ -n "$isminigc" ]; then
634 # Perform a "mini" gc
635 # Note that .delaygc is ignored here as that's only intended for full gc
637 rm -f .allowgc .needsgc
638 rm -f objects
/pack
/pack-
*_r.keep
644 if [ -f .svnpack
] && [ -n "$svn_mirror" ]; then
646 progress
"+ [$proj] mini garbage check ($(date))"
649 if [ -z "$cfg_delay_gfi_redelta" ] && [ -n "$gfi_mirror" ]; then
650 # $Girocco::Config::delay_gfi_redelta is false, force redeltification now
651 if [ -z "$miniactive" ]; then
653 progress
"+ [$proj] mini garbage check ($(date))"
657 # If there aren't at least 10 non-keep, non-bitmap, non-bndl packs then
658 # don't actually process them yet
659 lpo
="--exclude-no-idx --exclude-keep --exclude-bitmap --exclude-bndl --quiet"
660 packcnt
="$(list_packs --count $lpo objects/pack)" ||
:
661 if [ "${packcnt:-0}" -ge 10 ]; then
662 if [ -z "$miniactive" ]; then
664 progress
"+ [$proj] mini garbage check ($(date))"
666 if [ -n "$gfi_mirror" ]; then
668 packcnt
="$(list_packs --count $lpo objects/pack)" ||
:
670 # if repack_gfi_packs dropped the pack count to < 10 don't combine
671 if [ "${packcnt:-0}" -ge 10 ]; then
673 packcnt
="$(list_packs --count $lpo objects/pack)" ||
:
675 # if we still have more than 10 packs trigger a full gc
676 if [ "${packcnt:-0}" -ge 10 ]; then
677 # We shouldn't be in a .delaygc state at this point, but if
678 # we are then nuke it because we really need a full gc now
680 git config
--unset gitweb.lastgc
682 git update-server-info
# just in case
683 progress
"- [$proj] mini garbage check triggering full gc too many packs ($(date))"
688 if [ -n "$miniactive" ]; then
689 git update-server-info
690 progress
"- [$proj] mini garbage check ($(date))"
692 progress
"= [$proj] mini garbage check nothing but crud removal to do ($(date))"
697 # Avoid unnecessary garbage collections:
698 # 1. If lastreceive is set and is older than lastgc
700 # 2. We are not a fork (! -s alternates) -OR- lastparentgc is older than lastgc
702 # If lastgc is NOT set or lastreceive is NOT set we MUST run gc
703 # If we are a fork and lastparentgc is NOT set we MUST run gc
705 # If the repo is dirty after removing any crud we MUST run gc
707 gcstart
="$(date "$datefmt")"
710 ! [ -s objects
/info
/alternates
] || isfork
=1
712 [ -z "$isfork" ] || lastparentgcsecs
="$(config_get_date_seconds lastparentgc)" ||
:
714 if lastreceivesecs
="$(config_get_date_seconds lastreceive)" &&
715 [ "${force_gc:-0}" = "0" ] &&
716 lastgcsecs
="$(config_get_date_seconds lastgc)" &&
717 [ $lastreceivesecs -lt $lastgcsecs ]; then
718 # We've run gc since we last received, so maybe we can skip,
719 # check if not fork or fork and lastparentgc < lastgc
720 if [ -n "$isfork" ]; then
721 if [ -n "$lastparentgcsecs" ] &&
722 [ $lastparentgcsecs -lt $lastgcsecs ]; then
723 # We've run gc since our parent ran gc so we can skip
727 # We don't have any alternates (we're not a forK) so we can skip
732 # Prevent any other simultaneous gc operations
735 # At this point, if .allowgc or .gc_failed exists, it's now crud to be removed
736 rm -f .allowgc .gc_failed
738 # Ideally we would do this in post-receive, but that would mean duplicating the
739 # logic so it's available in the chroot jail and that's highly undesirable
740 # Instead, since the first gc will be triggered immediately following the first
741 # push, we do the check here as it's quick and harmless if HEAD is already valid
742 check_and_set_head ||
:
744 # Always get rid of crud
747 # Always perform reflogs maintenance
752 # Run 'git svn gc' now for svn mirrors
753 if [ -n "$svn_mirror" ]; then
757 # Skip the actual gc if .delaygc is set
758 if [ -e .delaygc
]; then
759 progress
"x [$proj] garbage check delayed (except for crud removal)"
764 # Do not skip gc if the repo is dirty
765 if [ -n "$skipgc" ] && ! is_dirty
; then
766 progress
"= [$proj] garbage check nothing but crud removal to do ($(date))"
767 config_set lastgc
"$gcstart"
773 if [ -n "$isfork" ] && [ -z "$lastparentgcsecs" ]; then
774 # set lastparentgc and then update gcstart to be at least 1 second later
775 config_set lastparentgc
"$gcstart"
778 if [ -z "$lastreceivesecs" ]; then
779 # set lastreceive and then update gcstart to be at least 1 second later
780 config_set lastreceive
"$gcstart"
783 if [ -n "$bumptime" ]; then
785 gcstart
="$(date "$datefmt")"
788 progress
"+ [$proj] garbage check ($(date))"
791 [ -z "$alwaysredelta" ] || newdeltas
="$noreusedeltaopt"
792 if [ -z "$newdeltas" ] && [ -n "$gfi_mirror" ]; then
793 if [ $
(list_packs
--exclude-no-idx --count objects
/pack
) -le \
794 $
(list_packs
--exclude-no-idx --count --quiet --only gfi-packs
) ]; then
795 # Don't bother with repack_gfi_packs since everything's being repacked
796 newdeltas
="--no-reuse-delta"
799 if [ -z "$newdeltas" ] && [ -n "$noreusedeltaopt" ] &&
800 [ $
(list_packs
--exclude-no-idx --count-objects objects
/pack
) -le $var_redelta_threshold ]; then
801 # There aren't enough objects to worry about so just redelta to get the best pack
802 newdeltas
="--no-reuse-delta"
804 if [ -z "$newdeltas" ]; then
805 # Since we're not going to recompute deltas overall, we need to do the
806 # "mini" maintenance so that we can get more optimal deltas
807 [ -z "$noreusedeltaopt" ] || make_svn_pack
809 force_single_pack_redelta
=
810 [ -n "$gfi_mirror" ] ||
[ -n "$svn_mirror" ] || force_single_pack_redelta
=1
811 [ -z "$noreusedeltaopt" ] || combine_small_packs
$force_single_pack_redelta
815 ## Safe Pruning In Forks
817 ## We are about to perform garbage collection. We do NOT use the "git gc" or
818 ## the "git repack" commands directly as they do not provide enough control over
819 ## the fine details. However, we DO maintain a "gc.pid" file during our garbage
820 ## collection so that a simultaneous "git gc" by an administrator will be
821 ## blocked (and similarly we refuse to start garbage collection if we cannot
822 ## create the "gc.pid" file).
824 ## When we say "gc" in the below description we are referring to our "gc.sh"
825 ## script, NOT the "git gc" command.
827 ## If the project we are running garbage collection (gc) on has any forks we
828 ## must be careful not to remove any objects that while no longer referenced by
829 ## this project (the parent) are still referenced by one or more forks (the
830 ## children) otherwise the children will become corrupt and we can't abide
833 ## One way to accomplish this is to simply hard-link all currently existing
834 ## loose objects and packs in the parent into all the children that refer to the
835 ## parent (via a line in their objects/info/alternates file) before beginning
836 ## the gc operation and then relying on a subsequent gc in the child to clean up
837 ## any excess objects/packs. We used to use this strategy but it's very
838 ## inefficient because:
840 ## 1. The disk space used by the old pack(s)/object(s) will not be reclaimed
841 ## until all children (and their children, if any) run gc by which time
842 ## it's quite possible the topmost parent will have run gc again and
843 ## hard-linked yet another old pack down to its children (not to mention
846 ## 2. When using the "-A" option with "git repack", any new objects in the
847 ## parent that are not referenced by children will continually get
848 ## exploded out of the hard-linked pack in the children whenever the
851 ## 3. To avoid suboptimal and/or unnecessarily many packs being hard-linked
852 ## into child forks, we must run the "mini" gc maintenance before we
853 ## perform the hard-linking into the children which provides yet another
854 ## source of inefficiency.
856 ## While we were still using the "-A" option to "git repack" (that was not
857 ## always the case) to guarantee we can access old ref values for long enough
858 ## to send out a meaningful mail.sh notification, another, more efficient,
859 ## option became available to prevent corruption of child forks that continue
860 ## to refer to objects that are no longer reachable from any ref in the parent.
862 ## The only things that need be copied (or hard-linked) into the child fork(s)
863 ## are those objects that have become unreachable from any ref in the parent.
865 ## When we were using the "git repack -A -d" + "git prune --expire=1.day.ago"
866 ## technique, the only objects that could ever be removed were loose objects
867 ## that "git prune" determined were expired. In that case, loose objects were
868 ## all that need be hard-linked down to child forks in order to avoid
869 ## corruption of any child fork(s).
871 ## The "git repack -A -d" + "git prune --expire=1.day.ago" + hard-linking loose
872 ## objects to child forks technique remains fundamentally sound from the
873 ## perspective of supporting simultaneous gc and push and keeping newly
874 ## unreachable objects around long enough to be sure we can send out meaningful
875 ## ref change notifications and never corrupting any child forks and never
876 ## persisting the lifetime of large old packs containing mostly duplicate or
877 ## unreachable objects as gc percolates through a project's entire fork tree.
879 ## However, that technique suffers from one potential prodigious pitfall.
881 ## Unreachable objects come flying out of their packs to splatter all over the
882 ## objects subdirectories possibly creating a huge, inefficient mess.
884 ## Often this is not an issue. Even with a lot of rebasing going on, usually
885 ## the only objects that will splatter are some commits, trees and the odd blob
886 ## here and there. Not enough to be overly concerned about.
888 ## However, for the reppository that frequently experiences a lot of non-fast-
889 ## forward updates and/or outright ref deletion, the number of objects suddenly
890 ## popping out of their packs at "git repack -A -d" time can be overwhelming.
892 ## To avoid this issue we now use a four phase pack creation strategy.
893 ## This will result in creation of up to four packs (instead of at most one).
895 ## I. A complete pack (with bitmaps if appropriate) gets created including
896 ## only "reachable" objects from all refs/... refs plus HEAD. This will
897 ## also serve as the virtual bundle for the repository.
899 ## II. A pack of recently-became-unreachable objects and friends is created.
900 ## (The "friends" are ref logs, linked working tree HEADs and indicies.)
901 ## Because both the pre-receive and update.sh script record all ref
902 ## changes we can easily choose the cut off point for "recently".
903 ## It is only the fact we maintain those logs in the reflogs subdirectory
904 ## that allows this step to be possible.
906 ## III. If the repository has any forks with a non-zero length alternates file,
907 ## yet another pack of "--keep-unreachable" objects is generated that will
908 ## not actually be kept in the parent, but hard-linked into all the forks.
910 ## IV. Finally, after running "git prune-packed", any remaining loose objects
911 ## are migrated into a pack of their own.
913 ## We then remove any non-.keep packs that existed before we started the
914 ## process being careful to keep any same-pack pushes for the "Push Pack Redux"
915 ## race condition (see README-GC).
917 ## By using "git pack-objects" directly we are able to accomplish this with
918 ## very little additional effort.
920 ## The packs produced by (III) are treated almost like ".keep" packs by child
921 ## forks in that the objects in them are never repacked into any other
922 ## "--keep-unreachable" packs (but they can migrate into phase I or II packs)
923 ## and those phase III packs are then hard-linked into any grandchild forks.
925 ## This avoids the space explosion that could occur if each fork level ended
926 ## up duplicating the "--keep-unreachable" pack space by repacking those
927 ## objects (essentially breaking the hard-link to the single copy of those
930 ## While it is true that each level of forks could potentially add yet another
931 ## phase III pack to be hard-linked down to its children, such packs will only
932 ## include unreachable objects not already in any phase III packs that were
933 ## received from the parent.
935 ## The space for the phase III packs will not be reclaimed until the gc
936 ## finishes percolating through the entire "fork tree" of a project.
938 ## This is not much different than the "git repack -A -d" situation where
939 ## all the loose objects are hard-linked down into child forks. In that
940 ## case forks that actually need any of those objects could gradually reduce
941 ## the number of objects hard-linked into deeper fork levels.
943 ## The difference with a phase III "--keep-unreachable" pack is that there
944 ## cannot be any gradual reduction like that since it would require repacking
945 ## the pack and breaking the hard-link thereby increasing storage space. The
946 ## storage will instead always be reclaimed all at once when all of the
947 ## projects in the "fork tree" complete their gc.
949 ## However, the belief is that the huge space win by having all the
950 ## unreachable objects packed up together far eclipses (when many objects are
951 ## involved, the single-pack version can end up using 1/20th or less of the
952 ## disk space compared to having them all as loose objects) any brief minor
953 ## space savings that might occur under the "git repack -A -d" loose object
954 ## system prior to the gc collection completing for all the projects in the
962 # rename_pack oldnamepath newnamepath
963 # note that .keep files are left untouched and not moved at all!
965 [ $# -eq 2 ] && [ "$1" != "$2" ] ||
{
966 echo >&2 "[$proj] incorrect use of rename_pack function"
969 # Git assumes that if the destination of the rename already exists
970 # that it is, in fact, a copy of the same bytes so silently succeeds
971 # without doing anything. We duplicate that logic here.
972 # Git checks for the .idx file first before even trying to use a pack
973 # so it should be the last moved and the first removed.
974 for ext
in pack bitmap idx
; do
975 [ -f "$1.$ext" ] ||
continue
976 ln "$1.$ext" "$2.$ext" >/dev
/null
2>&1 ||
977 [ -f "$2.$ext" ] ||
{
978 echo >&2 "[$proj] unable to move $1.$ext to $2.$ext"
982 for ext
in idx pack bitmap
; do
989 find -L "$1" -maxdepth 1 -type f
! -perm -ug+w \
990 -name "pack-$octet20*.pack" -exec chmod ug
+w
'{}' + ||
:
994 eval "$1="'$(( $# - 1 ))'
1000 export GIT_INDEX_FILE
1001 git write-tree
2>/dev
/null ||
:
1002 unset GIT_INDEX_FILE
1006 get_detached_head
() {
1007 if [ -s "$1" ] && read -r _head
<"$1" 2>/dev
/null
; then
1008 case "$_head" in $octet20*)
1014 # compute_extra_reachables
1015 # create lines suitable for a packed-refs file mentioning all the
1016 # other refs we might like to keep.
1017 # the current directory MUST be set to the repository's --git-dir
1018 # the following are included:
1019 # * refs mentioned in reflogs/... files
1020 # * tree(s) created from index file(s)
1021 # * detached linked working tree heads
1022 # Resulting objects are tested for existence and uniqified then output
1023 # one per line under a refs/z* namespace
1024 compute_extra_reachables
() {
1026 digits8
='[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]'
1027 find -L reflogs
-mindepth 1 -maxdepth 1 -type f
-name "$digits8*" -exec gzip -c -d -f '{}' + |
1028 awk '{print $2; print $3}'
1029 ! [ -f index
] || get_index_tree index
1030 if [ -d worktrees
]; then
1031 find -L worktrees
-mindepth 2 -maxdepth 2 -name HEAD
-type f
-print |
1032 while read -r lwth
; do
1033 get_detached_head
"$lwth"
1034 get_index_tree
"${lwth%HEAD}index"
1037 } | LC_ALL
=C
sort -u |
1038 git cat-file
${var_have_git_260:+--buffer} --batch-check"${var_have_git_185:+=%(objectname)}" |
1039 awk '!/missing/ {num++; print $1 " " "refs/" substr("zzzzzzzzzzzz", 1, length(num)) "/" num}'
1046 # Everything else is more efficient if we do this first
1047 # The "--prune" option is the default since v1.5.0 but it serves as "documentation" here
1048 git pack-refs
--all --prune
1049 [ -e packed-refs
] ||
>>packed-refs
# should never happen...
1051 # If we have a logs directory or a worktrees directory expire the ref logs now
1052 # Note that Git itself does not use either --rewrite or --updateref, so neither do we
1053 ! [ -d logs
] && ! [ -d worktrees
] ||
eval git reflog expire
--all "${quiet:+>/dev/null 2>&1}" ||
:
1056 ! [ -e .gc_failed
] ||
exit 1
1057 rm -f .gc_in_progress
# make sure
1058 touch .gc_in_progress
# it's truly fresh
1059 rm -f bundles
/* objects
/pack
/pack-
*.bndl
1060 # This is perhaps a bit aggressive in that if we're suffering from "Push Pack Redux"
1061 # and somehow we get run again immediately after the run where "Push Pack Redux" happened
1062 # and we have garbage collection forced, there's just the barest, almost negligible,
1063 # possibility that the "Push Pack Redux" ref updates _still_ have not happened and we
1064 # should not be removing _r .keep files. None of the normal Girocco processing can
1065 # cause this. The second run of this script would have to use the force gc option
1066 # for it to even be possible in the first place. What's much more likely is that
1067 # the initial run of this script was somehow interrupted in the middle before it
1068 # could get rid of the _r .keep file itself in which case it's better to get rid of
1069 # it now to avoid keeping something around that would perturb our nice and neat gc
1070 rm -f objects
/pack
/pack-
*_r.keep
1071 # We will add .keep files for _u packs if and when we run phase III
1072 # Otherwise they need to not have any .keep files during phases I and II
1073 rm -f objects
/pack
/pack-
*_u.keep
1075 # We need to make sure that any non-Girocco (barely tolerated) Git object creation
1076 # activity will be able to "freshen" the pack containing a pre-existing object
1077 # that's being written. This really should not be necessary as the pre-receive
1078 # hook should make sure this takes place for any incoming pushes.
1079 # However, do it here anyway just in case.
1080 make_packs_ugw objects
/pack
1082 # This is only effective with Git v2.3.5 and later and it will only matter when
1083 # we are using one of the "internal_rev_list" modes of pack-objects
1084 # (the combine-packs.sh script never uses any of those modes)
1085 # The "git repack" and "git prune" commands always sets this internally themselves
1086 # It makes no difference if there's no repository corruption
1087 GIT_REF_PARANOIA
=1 && export GIT_REF_PARANOIA
1089 # All of the options we might want to use with pack-objects were supported
1090 # at some point prior to Git version v1.6.6 which is the minimum version that
1091 # Girocco now requires. Except for one (--use-bitmap-index). Several of them
1092 # are "boiler plate" options we always want to use so we bundle them up here.
1093 pkopt
="--delta-base-offset --keep-true-parents --non-empty --all-progress-implied"
1094 # We want to use --include-tag, but before Git v2.10.1 it would leave out
1095 # "middle" tags (e.g. a tag of a tag of a commit would omit the tagged tag)
1096 # See http://repo.or.cz/git.git/b773ddea2cd3b08c for details
1097 # ("pack-objects: walk tag chains for --include-tag", 2016-09-07, v2.10.1)
1098 # This is not a free check as it matches all refs against refs/tags/ then
1099 # peels all the annotated tags and checks for inclusion. The situation in
1100 # which it would add a tag that was not already included by a reachability
1101 # trace that included tag starting points can only occur if a new tag gets
1102 # pushed during gc pointing to something that would have been packed anyway.
1103 # But, it could happen and, really, campared to gc as a whole it's not that
1104 # expensive to perform (provided we do not get an unconnected pack).
1105 [ -z "$var_have_git_2101" ] || pkopt
="$pkopt --include-tag"
1106 pkopt
="$pkopt ${quiet:---progress} $packopts"
1108 # The git pack-objects command only supports bitmaps if all objects are being
1109 # packed (the "--all" option) and the "--stdout" option is NOT being used.
1110 # Additionally, while packing, if any encountered reachable objects are
1111 # determined to be "not wanted" then no bitmap index will be written anyway.
1112 # While it is theoretically possible that a project with a non-empty alternates
1113 # file ends up packing all objects (because it does not actually use any of the
1114 # objects found in the alternates), it's very unlikely. And, in the unlikely
1115 # event that did occur, clients would see a message about only using one bitmap
1116 # because Git can only use one bitmap at a time and at least one of the
1117 # alternates is bound to have a bitmap. Therefore if we see a non-empty
1118 # alternates file, we disable writing bitmaps which avoids the warning and any
1119 # possibility of a client warning as well. Also if we are running anything
1120 # before Git v2.1.0 (the effective version for repack.writeBitmaps=true) then
1121 # we also always disable bitmap writing.
1123 [ -z "$var_have_git_210" ] || wbmopt
="--write-bitmap-index"
1124 # More recent versions of pack-objects have optimizations when not using the
1125 # --local option. If we do not have any alternates it's a pointless option.
1126 # If we do have alternates we need to skip writing a bitmap and we cannot
1127 # have a bundle since it must contain all objects.
1128 if [ -n "$isfork" ]; then
1142 [ -n "$wbmopt" ] || wbmstr
=" (bitmaps disabled)"
1143 progress
"~ [$proj] running primary full gc pack-objects$wbmstr"
1146 ! has_forks_with_alternates
"$proj" || gotforks
=1
1148 # To avoid "Push Pack Redux" (see README-GC), after collecting the initial
1149 # preexisting non-keep pack list, we rename them so that an incoming push
1150 # pack cannot possibly experience a pack name collision. Git does not require
1151 # use of the "default" pack names, simply that the proper extensions are used.
1152 # We rename to insert an "_r" just before the extension to avoid "Push Pack Redux"
1153 # name collisions. Later on we may create an "unreachable" pack for hard-linking
1154 # down into forks and it will have an "_u" inserted just before its extension.
1155 packlist
="$(list_packs -C objects/pack --exclude-no-idx --exclude-keep --quiet .)" ||
:
1157 for oldpack
in $packlist; do
1158 oldpack
="${oldpack#pack-}"
1159 oldpack
="${oldpack%.pack}"
1160 [ -f "objects/pack/pack-$oldpack.pack" ] ||
{
1161 echo >&2 "[$proj] unable to list old pack files"
1164 if [ "${oldpack#*[!0-9a-fA-F]}" != "$oldpack" ]; then
1165 # names not exclusively hexadecimal do not need renaming
1166 oldpacks
="${oldpacks:+$oldpacks }$oldpack"
1169 rename_pack
"objects/pack/pack-$oldpack" "objects/pack/pack-${oldpack}_r" ||
{
1170 echo >&2 "[$proj] unable to rename old pack files"
1173 # If the oldpack has a .keep now it means a "Push Pack Redux" is actually
1174 # in progress at this moment and we need to .keep the renamed pack,
1175 # otherwise no "Push Pack Redux" has started yet or it has already finished.
1176 # In either case we're okay because if it's just finished then all ref
1177 # changes have already been made so we don't need a .keep and we will
1178 # see the ref changes and grab all the objects via a reachability trace.
1179 # If it hasn't started yet that's okay because we're done moving that
1180 # name so a complete pack will appear under the old name that we'll
1182 if [ -f "objects/pack/pack-$oldpack.keep" ]; then
1183 echo "Push Pack Redux" >"objects/pack/pack-${oldpack}_r.keep"
1185 oldpacks
="${oldpacks:+$oldpacks }${oldpack}_r"
1189 # We wish to keep deltas from our last full pack so if we're not redeltaing
1190 # then make sure the .pack associated with the .bitmap has a newer mod time
1191 # (If there is no .bitmap then touch the pack with the most objects instead.)
1192 if [ -z "$newdeltas" ]; then
1193 bmpack
="$(list_packs --exclude-no-bitmap --exclude-no-idx --max-matches 1 objects/pack)"
1194 [ -n "$bmpack" ] || bmpack
="$(list_packs --exclude-no-idx --max-matches 1 --object-limit -1 --include-boundary objects/pack)"
1195 if [ -n "$bmpack" ] && [ -f "$bmpack" ] && [ -s "$bmpack" ]; then
1197 touch -c "$bmpack" 2>/dev
/null ||
:
1198 # We must touch .gc_in_progress here to avoid $bmpack looking
1199 # like it's been "freshened" when redundant packs are removed
1200 # It's okay if they have the same mod time, but POSIX does not
1201 # guarantee an ordering for the "touching" that occurs which is
1202 # why this must be a separate command but needs no "sleep 1"
1203 touch .gc_in_progress
1207 # Now we need to make sure that any "freshening" that takes place will actually
1208 # result in a "newer" modification time than the .gc_in_progress file now has
1211 # We run git pack-objects from the repack subdirectory so we can force
1212 # optimized packs to be generated even for repositories that do not have any
1214 packs
="$(git --git-dir=repack pack-objects </dev/null \
1215 $pkopt --all $newdeltas $lclopt ${wbmopt:---honor-pack-keep} $@ repack/alt/pack/pack)"
1217 [ $packcnt -eq 1 ] || makebndl
=
1223 progress
"~ [$proj] running supplementary gc pack-objects"
1225 # Add the "supplementary" refs
1226 compute_extra_reachables
>>repack
/packed-refs
1228 # Subtract the primary refs
1229 GIT_ALTERNATE_OBJECT_DIRECTORIES
="$PWD/repack/alt"
1230 export GIT_ALTERNATE_OBJECT_DIRECTORIES
1232 # For this one we MUST use --local and MUST NOT use --write-bitmap-index
1233 # However, if there is a "logs" subdirectory we need to use --reflog
1234 # We do add it, just in case, if the linked working trees dir is present
1235 # We do not add --indexed-objects as that requires v2.2.0 and it's unclear
1236 # if it properly includes linked working tree index files or not. The
1237 # above compute_extra_reachables has already included all index trees (thereby
1238 # providing proper --indexed-objects support for all Git versions) making the
1239 # option completely unnecessary.
1241 ! [ -d logs
] && ! [ -d worktrees
] || rflopt
=--reflog
1242 spacks
="$(git --git-dir=repack pack-objects </dev/null \
1243 $pkopt --honor-pack-keep --all $rflopt $newdeltas --local $@ repack/alt/pack/pack)"
1249 # There's nothing to do for Phase III unless we have forks that refer to our
1250 # project from their alternates file
1253 if [ -n "$gotforks" ]; then
1255 progress
"~ [$proj] running keep-unreachable gc pack-objects for forks"
1257 # If we are a fork, any pre-existing _u packs need to have a .keep
1258 # for this phase and be added to the hlpacks list otherwise (we are
1259 # not a fork) pre-existing _u packs are anomalies to be treated like
1260 # regular non-_u packs
1261 if [ -n "$isfork" ]; then
1262 for upack
in $
(find -L objects
/pack
-mindepth 1 -maxdepth 1 -name "pack-$octet20*_u.pack" -print); do
1263 upack
="${upack%.pack}"
1264 [ -e "$upack.keep" ] ||
echo "unreachable" >"$upack.keep"
1265 hlpacks
="${hlpacks:+$hlpacks }${upack#objects/pack/pack-}"
1268 # Using either --no-reuse-delta or --no-reuse-object together with the
1269 # --keep-unreachable option is a very, very, very bad idea when good
1270 # packs are the desired outcome. If newdeltas are being generated
1271 # then we pack to a temp name, and use combine-packs.sh to get a better
1272 # pack as the result to avoid making a bad --keep-unreachable pack
1274 [ -z "$newdeltas" ] || pfx
="ku"
1275 upacks
="$(git --git-dir=repack pack-objects </dev/null \
1276 $pkopt --honor-pack-keep --all $rflopt --keep-unreachable --local $@ repack/alt/pack/${pfx}pack)"
1277 if [ -n "$upacks" ] && [ -n "$newdeltas" ]; then
1278 progress
"~ [$proj] rebuilding keep-unreachable pack deltas"
1281 printf "repack
/alt
/pack
/${pfx}pack-
%s.pack
\n" $oldupacks |
1282 run_combine_packs --names --weak-naming --non-empty --all-progress-implied ${quiet:---progress} \
1283 $packopts $newdeltas $@ repack/alt/pack/pack)"
1284 eval rm -f "$(printf \""repack
/alt
/pack
/${pfx}pack-
%s.
*"\"" " $oldupacks)"
1286 for upack
in $upacks; do
1287 rename_pack
"repack/alt/pack/pack-$upack" "repack/alt/pack/pack-${upack}_u"
1289 rm -f objects
/pack
/pack-
*_u.keep
1290 [ -z "$hlpacks" ] && [ -z "$upacks" ] ||
1291 progress
"~ [$proj] hard-linking keep-unreachable pack(s) into immediate child forks"
1293 # We have to update the lastparentgc time in the child forks even if they do not get any
1294 # new "unreachable packs" because they need to run gc just in case the parent now has some
1295 # objects that used to only be in the child so they can be removed from the child.
1296 # For example, a "patch" might be developed first in a fork and then later accepted into
1297 # the parent in which case the objects making up the patch in the child fork are now
1298 # redundant (since they're now in the parent as well) and need to be removed from the
1299 # child fork which can only happen if the child fork runs gc.
1300 lastparentgc
="$(date "$datefmt")"
1302 # It is enough to copy objects just one level down and get_repo_list
1303 # takes a regular expression (which is automatically prefixed with '^')
1304 # so we can easily match forks exactly one level down from this project
1306 get_repo_list
"$forkdir/[^/:][^/:]*:" |
1308 # Ignore forks that do not exist or are symbolic links
1309 ! [ -L "$cfg_reporoot/$fork.git" ] && [ -d "$cfg_reporoot/$fork.git" ] ||
1311 # Or do not have a non-zero length alternates file
1312 [ -s "$cfg_reporoot/$fork.git/objects/info/alternates" ] ||
1315 # Match hlpacks in parent project if any
1316 if [ -n "$hlpacks" ]; then
1317 mkdir
-p "$cfg_reporoot/$fork.git/objects/pack"
1318 eval ln -f "$(printf '"objects
/pack
/pack-
%s.pack
" ' $hlpacks)" \
1319 "$(printf '"objects
/pack
/pack-
%s.idx
" ' $hlpacks)" \
1320 '"$cfg_reporoot/$fork.git/objects/pack/"'
1323 # Match upacks in repack/alt area if any
1324 if [ -n "$upacks" ]; then
1325 mkdir
-p "$cfg_reporoot/$fork.git/objects/pack"
1326 eval ln -f "$(printf '"repack
/alt
/pack
/pack-
%s_u.pack
" ' $upacks)" \
1327 "$(printf '"repack
/alt
/pack
/pack-
%s_u.idx
" ' $upacks)" \
1328 '"$cfg_reporoot/$fork.git/objects/pack/"'
1331 if ! [ -e "$cfg_reporoot/$fork.git/.needsgc" ]; then
1332 # Trigger a mini gc in the fork if it now has too many packs
1333 packs
="$(list_packs --quiet --count --exclude-no-idx --exclude-keep "$cfg_reporoot/$fork.git
/objects
/pack
")" ||
:
1334 if [ -n "$packs" ] && [ "$packs" -ge 20 ]; then
1335 >"$cfg_reporoot/$fork.git/.needsgc"
1338 [ -z "$runupdate" ] || git
--git-dir="$cfg_reporoot/$fork.git" update-server-info
1339 # Update the fork's lastparentgc date (must be more recent than $gcstart)
1340 git
--git-dir="$cfg_reporoot/$fork.git" config gitweb.lastparentgc
"$lastparentgc"
1344 # Now move any primary/supplementary packs back into objects/pack
1345 # then drop any "unfreshened" redundant packs and clear repack/alt
1347 # First make sure the primary pack(s) have the most recent mod time
1348 [ -z "$packs" ] ||
printf 'repack/alt/pack/pack-%s.pack\n' $packs |
xargs touch -c 2>/dev
/null ||
:
1350 # Move the packs into place
1351 for pack
in $packs $spacks; do
1352 rename_pack
"repack/alt/pack/pack-$pack" "objects/pack/pack-$pack"
1355 # It's possible that one of the $oldpacks had a .bitmap, got renamed (along
1356 # with its .bitmap) and then got "freshened" causing us to not remove it
1357 # However, if $wbmopt is set we most likely now have TWO .bitmap packs!
1358 # This can produce ugly warnings we don't want and possibly get the wrong
1359 # bitmap used since only one .bitmap file can ever be used by Git.
1360 # If this has happened, the .bitmap we want to discard will always have
1361 # an _r infix so we can just zap any such now since it will leave the pack.
1362 [ -z "$wbmopt" ] ||
rm -f objects
/pack
/pack-
*_r.bitmap ||
:
1364 # Remove the redundant packs that have not since been "freshened"
1365 # This does not completely eliminate the race condition window (Girocco's own
1366 # activites -- gc/fetch/receive are immune to the race) but it substantially
1367 # shrinks it down to just the time after the find but before the following rm
1369 [ -z "$oldpacks" ] ||
1370 printf 'objects/pack/pack-%s.pack\n' $oldpacks |
1371 LC_ALL
=C
sort >repack
/oldpacks
1372 find -L objects
/pack
-maxdepth 1 -type f
-name "pack-$octet20*.pack" -newer .gc_in_progress
-print |
1373 LC_ALL
=C
sort >repack
/freshened
1374 deadpacks
="$(LC_ALL=C join -v 1 repack/oldpacks repack/freshened | LC_ALL=C sed 's/\.pack$//')"
1375 [ -z "$deadpacks" ] ||
1376 eval echo "$(printf '"%s
".* ' $deadpacks)" |
xargs rm -f ||
:
1378 # No need for this anymore
1379 rm -rf repack
/alt objects
/pack
/repack
1380 unset GIT_ALTERNATE_OBJECT_DIRECTORIES
1386 progress
"~ [$proj] running gc prune-packed"
1388 # We do not want the redundant packs or any new "--keep-unreachable" pack(s) to be
1389 # present while running prune-packed. We try to guarantee that any loose object
1390 # that's unreachable persists for at least one $Girocco::Config::min_gc_interval
1391 # (not withstanding administrator interference to force earlier gc to occur).
1392 # If we were to include the redundant/keep-unreachable pack(s) when running
1393 # prune-packed and a loose unreachable object happened to be duplicated in one
1394 # of them we would end up removing it too soon and void our guarantee.
1395 git prune-packed
$quiet
1397 progress
"~ [$proj] running loose objects gc pack-objects"
1399 # Although Git v2.10.0 and later support a --pack-loose-unreachable option,
1400 # we MUST NOT use it for these reasons:
1401 # 1) We're not interested in expensive "unreachable" at this point, only "loose"
1402 # 2) It produces simply horrid packs about 3.8x times larger than they should be
1403 # 3) We don't require anything more than Git v1.6.6
1404 lpacks
="$(run_combine_packs </dev/null --names --loose --weak-naming --non-empty --all-progress-implied ${quiet:---progress} $packopts $newdeltas $@)"
1406 if [ -n "$lpacks" ]; then
1407 # Make sure any primary pack(s) have a more recent mod time than "unreachable" objects packs
1408 if [ -n "$packs" ]; then
1410 printf 'objects/pack/pack-%s.pack\n' $packs |
xargs touch -c 2>/dev
/null ||
:
1412 # We need to identify these packs later so we don't combine_packs them
1413 for objpack
in $lpacks; do
1414 rename_pack
"objects/pack/pack-$objpack" "objects/pack/pack-${objpack}_o" ||
:
1416 # Finally zap the corresponding loose objects
1417 progress
"~ [$proj] running packed loose objects gc prune-packed"
1418 git prune-packed
$quiet
1421 ! [ -e .gc_failed
] ||
exit 1
1422 # These, if they exist, are now meaningless and need to be removed
1423 rm -f gfi-packs .needsgc .svnpack .svnpackgc
1425 # Make sure this stays up to date
1426 git update-server-info
1428 # We must make loose objects group writable so that they
1429 # can be freshened by other pushers. Technically we need only do this for
1430 # push projects but to enable mirror projects to be more easily converted to
1431 # push projects, we go ahead and do it for all projects.
1432 # By the time we get here we really shouldn't have any of these, but just in case.
1433 { find -L objects
/$octet -type f
-name "$octet19*" -exec chmod ug
+w
'{}' + ||
:; } 2>/dev
/null
1435 # darcs:// mirrors have a xxx.log file that will grow endlessly
1436 # if this is a mirror and the file exists, shorten it to 10000 lines
1437 # also take this opportunity to optimize the darcs repo
1438 if ! [ -e .nofetch
] && [ -n "$cfg_mirror" ]; then
1439 url
="$(config_get baseurl)" ||
:
1440 case "$url" in darcs
://*)
1441 if [ -n "$cfg_mirror_darcs" ]; then
1443 basedarcs
="$(basename "${url#darcs:/}")"
1444 if [ -f "$basedarcs.log" ]; then
1445 tail -n 10000 "$basedarcs.log" >"$basedarcs.log.$$"
1446 mv -f "$basedarcs.log.$$" "$basedarcs.log"
1448 if [ -d "$basedarcs.darcs" ]; then
1450 cd "$basedarcs.darcs"
1451 # without show_progress suppress non-error output
1452 [ -n "$show_progress" ] ||
exec >/dev
/null
1453 # Note that this does not optimize _darcs/inventories/ :(
1461 # Create a matching .bndl header file for the all-in-one pack we just created
1462 # but only if we're not a fork (otherwise the bundle would not be complete)
1463 # and we are running at least Git version 1.7.2 (pack_is_complete always fails otherwise)
1464 if [ -n "$makebndl" ] && [ -n "$var_have_git_172" ]; then
1465 # There should only be one pack in $packs but do some checking...
1466 # The one we just created will have a .idx and will NOT have a .keep
1467 progress
"~ [$proj] creating downloadble bundle header"
1470 IFS
= read -r curhead
<repack
/HEAD.orig ||
:
1472 [ -s "objects/pack/pack-$packs.pack" ] &&
1473 [ -s "objects/pack/pack-$packs.idx" ] &&
1474 ! [ -e "objects/pack/pack-$packs.keep" ] &&
1475 pkhead
="$(pack_is_complete "$PWD/objects
/pack
/pack-
$packs.pack
" \
1476 "$PWD/repack
/packed-refs.orig
" "$curhead")"
1478 pkbase
="objects/pack/pack-$packs"
1480 if [ -n "$pkbase" ] && [ -n "$pkhead" ]; then
1483 case "$curhead" in "ref: refs/"?
*|
"ref:refs/"?
*|
"refs/"?
*)
1484 symref
="${curhead#ref:}"
1485 symref
="${symref# }"
1488 [ -z "$cfg_httpbundleurl" ] || bndlurl
=" url=$cfg_httpbundleurl/$proj.git/clone.bundle"
1489 echo "# v2 git bundle"
1490 LC_ALL
=C
sed -ne "/^$octet20$hexdig* refs\/[^ $tab]*\$/ p" <repack
/packed-refs.orig
1491 if [ -n "$symref" ]; then
1492 printf "$pkhead HEAD\0symref=HEAD:%s%s\n" "$symref" "$bndlurl"
1494 if [ -n "$bndlurl" ]; then
1495 printf "$pkhead HEAD\0%s\n" "${bndlurl:# }"
1502 bndletag
="$("$cfg_basedir/bin
/rangecgi
" --etag -m 1 "$pkbase.bndl
" "$pkbase.pack
")" ||
:
1503 bndlsha
="$(printf '%s' "$bndletag" | git hash-object --stdin)" ||
:
1504 if [ -n "$bndletag" ]; then
1505 case "$bndlsha" in $octet20*)
1506 bndlshatrailer
="${bndlsha#????????}"
1507 bndlshaprefix
="${bndlsha%$bndlshatrailer}"
1508 bndlname
="$(TZ=UTC date +%Y%m%d_%H%M%S)-${bndlshaprefix:-0}"
1509 [ -d bundles
] || mkdir bundles
1510 echo "${pkbase#objects/pack/}.bndl" >"bundles/$bndlname"
1511 echo "${pkbase#objects/pack/}.pack" >>"bundles/$bndlname"
1512 ln -s -f -n "$bndlname" bundles
/latest
1518 # Record the size of this repo as the sum of its clone packed-refs + *.pack sizes as 1024-byte blocks
1519 eval "reposizek=$(( $(
1520 echo 0 $(du -k repack/packed-refs.orig $(printf 'objects/pack/pack-%s.pack ' $packs) 2>/dev/null |
1521 LC_ALL=C awk '{print $1}') |
1522 LC_ALL=C sed -e 's/ / + /g') ))"
1523 config_set_raw girocco.reposizek
"${reposizek:-0}"
1525 # Now we're finally done with this
1528 # We didn't used to do anything about rerere or worktrees but we're
1529 # trying to make nice with linked working trees these days :)
1530 # Maybe even non-bare repositories too, but *shush* about those ;)
1531 if [ -n "$var_have_git_250" ] && [ -d worktrees
]; then
1532 # The value "3.months.ago" is hard-coded into gc.c rather than
1533 # having the default be in worktree.c so we must provide it if
1534 # we get nothing out of the gc.worktreePruneExpire config item
1535 # Prior to Git v2.6.0 the config item was gc.pruneworktreesexpire
1536 # however we just always use the newer name no matter what Git version
1537 expiry
="$(git config --get gc.worktreePruneExpire 2>/dev/null)" ||
:
1538 eval git worktree prune
--expire '"${expiry:-3.months.ago}"' "${quiet:+>/dev/null 2>&1}" ||
:
1540 # git rerere does it right and handles its own default/config'd expiration values
1541 ! [ -d rr-cache
] ||
eval git rerere gc
"${quiet:+>/dev/null 2>&1}" ||
:
1543 # We use $gcstart here to avoid a race where a push occurs during the gc itself
1544 # and the next future gc could be incorrectly skipped if we used the current
1545 # timestamp here instead
1546 config_set lastgc
"$gcstart"
1549 progress
"- [$proj] garbage check ($(date))"