Commit 5981ca03 by Marco Mariani

gzip: added --rsyncable option; kvm: compress backup images

1 parent a6456dec
......@@ -10,3 +10,12 @@ url = ftp://ftp.gnu.org/pub/gnu/gzip/gzip-1.6.tar.xz
md5sum = da981f86677d58a106496e68de6f8995
environment =
PATH=${xz-utils:location}/bin:%(PATH)s
patch-options = -p1
# The --rsyncable patch is from debian/ubuntu,
# specifically https://launchpad.net/ubuntu/+source/gzip/1.6-3ubuntu1
# It is required to minimize the bandwidth used by rsync.
# For an explanation, see http://beeznest.wordpress.com/2005/02/03/rsyncable-gzip/
# Hunks for .texi files have been removed to avoid a dependency on makeinfo.
patches =
${:_profile_base_location_}/rsyncable.diff#0587af03a5580e2b7b4007469ee2b601
--- a/deflate.c
+++ b/deflate.c
@@ -131,6 +131,14 @@
#endif
/* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
+#ifndef RSYNC_WIN
+# define RSYNC_WIN 8192
+#endif
+/* Size of rsync window, must be < MAX_DIST */
+
+#define RSYNC_SUM_MATCH(sum) (((sum) & (RSYNC_WIN - 1)) == 0)
+/* Whether window sum matches magic value */
+
/* ===========================================================================
* Local data used by the "longest match" routines.
*/
@@ -212,6 +220,8 @@
unsigned good_match;
/* Use a faster search when the previous match is longer than this */
+local ulg rsync_sum; /* rolling sum of rsync window */
+local ulg rsync_chunk_end; /* next rsync sequence point */
/* Values for max_lazy_match, good_match and max_chain_length, depending on
* the desired pack level (0..9). The values given below have been tuned to
@@ -314,6 +324,10 @@
#endif
/* prev will be initialized on the fly */
+ /* rsync params */
+ rsync_chunk_end = 0xFFFFFFFFUL;
+ rsync_sum = 0;
+
/* Set the default configuration parameters:
*/
max_lazy_match = configuration_table[pack_level].max_lazy;
@@ -550,6 +564,8 @@
memcpy((char*)window, (char*)window+WSIZE, (unsigned)WSIZE);
match_start -= WSIZE;
strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */
+ if (rsync_chunk_end != 0xFFFFFFFFUL)
+ rsync_chunk_end -= WSIZE;
block_start -= (long) WSIZE;
@@ -579,13 +595,44 @@
}
}
+local void rsync_roll(unsigned start, unsigned num)
+{
+ unsigned i;
+
+ if (start < RSYNC_WIN) {
+ /* before window fills. */
+ for (i = start; i < RSYNC_WIN; i++) {
+ if (i == start + num) return;
+ rsync_sum += (ulg)window[i];
+ }
+ num -= (RSYNC_WIN - start);
+ start = RSYNC_WIN;
+ }
+
+ /* buffer after window full */
+ for (i = start; i < start+num; i++) {
+ /* New character in */
+ rsync_sum += (ulg)window[i];
+ /* Old character out */
+ rsync_sum -= (ulg)window[i - RSYNC_WIN];
+ if (rsync_chunk_end == 0xFFFFFFFFUL && RSYNC_SUM_MATCH(rsync_sum))
+ rsync_chunk_end = i;
+ }
+}
+
+/* ===========================================================================
+ * Set rsync_chunk_end if window sum matches magic value.
+ */
+#define RSYNC_ROLL(s, n) \
+ do { if (rsync) rsync_roll((s), (n)); } while(0)
+
/* ===========================================================================
* Flush the current block, with given end-of-file flag.
* IN assertion: strstart is set to the end of the current match.
*/
#define FLUSH_BLOCK(eof) \
flush_block(block_start >= 0L ? (char*)&window[(unsigned)block_start] : \
- (char*)NULL, (long)strstart - block_start, (eof))
+ (char*)NULL, (long)strstart - block_start, flush-1, (eof))
/* ===========================================================================
* Processes a new input file and return its compressed length. This
@@ -596,7 +643,7 @@
local off_t deflate_fast()
{
IPos hash_head; /* head of the hash chain */
- int flush; /* set if current block must be flushed */
+ int flush = 0; /* set if current block must be flushed, 2=>and padded */
unsigned match_length = 0; /* length of best match */
prev_length = MIN_MATCH-1;
@@ -626,7 +673,8 @@
lookahead -= match_length;
- /* Insert new strings in the hash table only if the match length
+ RSYNC_ROLL(strstart, match_length);
+ /* Insert new strings in the hash table only if the match length
* is not too large. This saves time but degrades compression.
*/
if (match_length <= max_insert_length) {
@@ -654,9 +702,14 @@
/* No match, output a literal byte */
Tracevv((stderr,"%c",window[strstart]));
flush = ct_tally (0, window[strstart]);
+ RSYNC_ROLL(strstart, 1);
lookahead--;
strstart++;
}
+ if (rsync && strstart > rsync_chunk_end) {
+ flush = 2;
+ rsync_chunk_end = 0xFFFFFFFFUL;
+ }
if (flush) FLUSH_BLOCK(0), block_start = strstart;
/* Make sure that we always have enough lookahead, except
@@ -679,7 +732,7 @@
{
IPos hash_head; /* head of hash chain */
IPos prev_match; /* previous match */
- int flush; /* set if current block must be flushed */
+ int flush = 0; /* set if current block must be flushed */
int match_available = 0; /* set if previous match exists */
register unsigned match_length = MIN_MATCH-1; /* length of best match */
@@ -730,6 +783,7 @@
*/
lookahead -= prev_length-1;
prev_length -= 2;
+ RSYNC_ROLL(strstart, prev_length+1);
do {
strstart++;
INSERT_STRING(strstart, hash_head);
@@ -742,24 +796,38 @@
match_available = 0;
match_length = MIN_MATCH-1;
strstart++;
- if (flush) FLUSH_BLOCK(0), block_start = strstart;
+ if (rsync && strstart > rsync_chunk_end) {
+ rsync_chunk_end = 0xFFFFFFFFUL;
+ flush = 2;
+ }
+ if (flush) FLUSH_BLOCK(0), block_start = strstart;
} else if (match_available) {
/* If there was no match at the previous position, output a
* single literal. If there was a match but the current match
* is longer, truncate the previous match to a single literal.
*/
Tracevv((stderr,"%c",window[strstart-1]));
- if (ct_tally (0, window[strstart-1])) {
- FLUSH_BLOCK(0), block_start = strstart;
- }
+ flush = ct_tally (0, window[strstart-1]);
+ if (rsync && strstart > rsync_chunk_end) {
+ rsync_chunk_end = 0xFFFFFFFFUL;
+ flush = 2;
+ }
+ if (flush) FLUSH_BLOCK(0), block_start = strstart;
+ RSYNC_ROLL(strstart, 1);
strstart++;
lookahead--;
} else {
/* There is no previous match to compare with, wait for
* the next step to decide.
*/
+ if (rsync && strstart > rsync_chunk_end) {
+ rsync_chunk_end = 0xFFFFFFFFUL;
+ flush = 2;
+ FLUSH_BLOCK(0), block_start = strstart;
+ }
match_available = 1;
+ RSYNC_ROLL(strstart, 1);
strstart++;
lookahead--;
}
--- a/gzip.c
+++ b/gzip.c
@@ -215,6 +215,7 @@
unsigned insize; /* valid bytes in inbuf */
unsigned inptr; /* index of next byte to be processed in inbuf */
unsigned outcnt; /* bytes in output buffer */
+int rsync = 0; /* make ryncable chunks */
static int handled_sig[] =
{
@@ -275,7 +276,7 @@
{"best", 0, 0, '9'}, /* compress better */
{"lzw", 0, 0, 'Z'}, /* make output compatible with old compress */
{"bits", 1, 0, 'b'}, /* max number of bits per code (implies -Z) */
-
+ {"rsyncable", 0, 0, 'R'}, /* make rsync-friendly archive */
{ 0, 0, 0, 0 }
};
@@ -359,6 +360,7 @@
" -Z, --lzw produce output compatible with old compress",
" -b, --bits=BITS max number of bits per code (implies -Z)",
#endif
+ " --rsyncable Make rsync-friendly archive",
"",
"With no FILE, or when FILE is -, read standard input.",
"",
@@ -489,8 +491,11 @@
#else
recursive = 1;
#endif
- break;
- case 'S':
+ break;
+ case 'R':
+ rsync = 1; break;
+
+ case 'S':
#ifdef NO_MULTIPLE_DOTS
if (*optarg == '.') optarg++;
#endif
--- a/gzip.h
+++ b/gzip.h
@@ -140,6 +140,7 @@
extern unsigned insize; /* valid bytes in inbuf */
extern unsigned inptr; /* index of next byte to be processed in inbuf */
extern unsigned outcnt; /* bytes in output buffer */
+extern int rsync; /* deflate into rsyncable chunks */
extern off_t bytes_in; /* number of input bytes */
extern off_t bytes_out; /* number of output bytes */
@@ -287,7 +288,7 @@
/* in trees.c */
extern void ct_init (ush *attr, int *method);
extern int ct_tally (int dist, int lc);
-extern off_t flush_block (char *buf, ulg stored_len, int eof);
+extern off_t flush_block (char *buf, ulg stored_len, int pad, int eof);
/* in bits.c */
extern void bi_init (file_t zipfile);
--- a/gzip.1
+++ b/gzip.1
@@ -5,6 +5,7 @@
.ll +8
.B gzip
.RB [ " \-acdfhklLnNrtvV19 " ]
+.RB [ --rsyncable ]
.RB [ \-S\ suffix ]
[
.I "name \&..."
@@ -287,6 +288,16 @@
.I gunzip
).
.TP
+.B --rsyncable
+While compressing, synchronize the output occasionally based on the input.
+This increases size by less than 1 percent most cases, but means that the
+.BR rsync (1)
+program can take advantage of similarities in the uncompressed input
+when syncronizing two files compressed with this flag.
+.I gunzip
+cannot tell the difference between a compressed file created with this option,
+and one created without it.
+.TP
.B \-S .suf --suffix .suf
When compressing, use suffix .suf instead of .gz.
Any non-empty suffix can be given, but suffixes
--- a/trees.c
+++ b/trees.c
@@ -856,9 +856,10 @@
* trees or store, and output the encoded block to the zip file. This function
* returns the total compressed length for the file so far.
*/
-off_t flush_block(buf, stored_len, eof)
+off_t flush_block(buf, stored_len, pad, eof)
char *buf; /* input block, or NULL if too old */
ulg stored_len; /* length of input block */
+ int pad; /* pad output to byte boundary */
int eof; /* true if this is the last block for a file */
{
ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */
@@ -951,6 +952,10 @@
Assert (input_len == bytes_in, "bad input size");
bi_windup();
compressed_len += 7; /* align on byte boundary */
+ } else if (pad && (compressed_len % 8) != 0) {
+ send_bits((STORED_BLOCK<<1)+eof, 3); /* send block type */
+ compressed_len = (compressed_len + 3 + 7) & ~7L;
+ copy_block(buf, 0, 1); /* with header */
}
return compressed_len >> 3;
......@@ -84,7 +84,7 @@ command =
[template]
recipe = slapos.recipe.template
url = ${:_profile_base_location_}/instance.cfg.in
md5sum = bc5a986c7208d02d3284a897ea90b39d
md5sum = 4c8f07da2217e54163c265fe6fe3d41d
output = ${buildout:directory}/template.cfg
mode = 0644
......@@ -115,7 +115,7 @@ on-update = true
[template-kvm-import]
recipe = slapos.recipe.template
url = ${:_profile_base_location_}/instance-kvm-import.cfg.in
md5sum = 98b1acc438895c0fa2309144a9a52a62
md5sum = 6835c9309ff4bf4a0efd1850e6c66b24
output = ${buildout:directory}/template-kvm-import.cfg
mode = 0644
......@@ -123,7 +123,7 @@ mode = 0644
recipe = hexagonit.recipe.download
url = ${:_profile_base_location_}/template/kvm-import.sh.jinja2
filename = kvm-import.sh.jinja2
md5sum = 2178e7ef7f1b9ea6639bfbaf045e1feb
md5sum = 926a11421921c29f91fae8240bbcf585
download-only = true
mode = 0755
......@@ -131,7 +131,7 @@ mode = 0755
recipe = hexagonit.recipe.download
url = ${:_profile_base_location_}/instance-kvm-export.cfg.jinja2
mode = 644
md5sum = 4972bc6f6a4cb41c1820ac13a32915a8
md5sum = c9f13c1f481ed08c75089aef1d3c6981
download-only = true
on-update = true
......@@ -139,7 +139,7 @@ on-update = true
recipe = hexagonit.recipe.download
url = ${:_profile_base_location_}/template/kvm-export.sh.jinja2
filename = kvm-export.sh.jinja2
md5sum = 2e2a99e33e5cdde35833d164da1e325b
md5sum = 22bd2e0c8fdb39a764a14c403a3bd752
download-only = true
mode = 0755
......
......@@ -27,6 +27,7 @@ context =
section directory directory
section buildout buildout
key socket_path kvm-instance:socket-path
raw gzip_binary {{ gzip_binary }}
# Extends publish section with resilient parameters
[publish-connection-information]
......
......@@ -33,5 +33,7 @@ mode = 0700
wrapper = $${:rendered}
context =
section directory directory
raw zcat_binary ${gzip:location}/bin/zcat
raw gzip_binary ${gzip:location}/bin/gzip
backup-disk-path = $${directory:backup}/virtual.qcow2
......@@ -82,6 +82,7 @@ context =
raw kvm_template $${dynamic-template-kvm:rendered}
raw template_kvm_export ${template-kvm-export-script:location}/${template-kvm-export-script:filename}
raw pbsready_export_template ${pbsready-export:output}
raw gzip_binary ${gzip:location}/bin/gzip
mode = 0644
[dynamic-template-kvm-resilient-test]
......
......@@ -11,5 +11,19 @@ QMP_CLIENT={{ buildout['directory'] }}/software_release/bin/qemu-qmp-client
$QMP_CLIENT --socket {{ socket_path }} --drive-backup $BACKUP_DIR/$BACKUP_FILE
# Due to the way qmp works, the VM file cannot be compressed on the fly.
# Although the compression step is optional, the importer uses the .gz file
# if present. So, remove it if you are disabling the compression.
# The downside of compression, here, is the temporary usage of more disk space
# in the exporter node. The goal is to minimize disk usage on the PBS server.
# If you want to compress the file in-place:
# truncate -s $(gzip -c $BACKUP_FILE | dd of=$BACKUP_FILE conv=notrunc 2>&1 | sed -n '$ s/ .*$// p') $BACKUP_FILE
# but the importer script would have to be adapted.
echo "Compressing backup..."
{{ gzip_binary }} --force --rsyncable $BACKUP_DIR/$BACKUP_FILE
cd $BACKUP_DIR && find -type f ! -name backup.signature -print0 | xargs -P4 -0 sha256sum | LC_ALL=C sort -k 66 > backup.signature
......@@ -16,7 +16,11 @@ write_backup_proof () {
# For now we just make the diff before
write_backup_proof
# TODO: Use rdiff
rm $VM_DIR/$VM_FILE
cp $BACKUP_DIR/$VM_FILE $VM_DIR/$VM_FILE
if [ -f "$BACKUP_DIR/${VM_FILE}.gz" ]; then
{{ gzip_binary }} -t "$BACKUP_DIR/${VM_FILE}.gz" || exit 10
{{ zcat_binary }} "$BACKUP_DIR/${VM_FILE}.gz" > $VM_DIR/$VM_FILE
else
rm $VM_DIR/$VM_FILE
cp $BACKUP_DIR/$VM_FILE $VM_DIR/$VM_FILE
fi
Styling with Markdown is supported
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!