Skip to content
extract random positioned data from gzip files with no penalty, including gzip tailing like with 'tail -f' !
C
Branch: master
Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
README.md
gztool.c note on readme about line number counting (-x) Dec 26, 2019

README.md

gztool

GZIP files indexer, compressor and data retriever. Create small indexes for gzipped files and use them for quick and random data extraction. No more waiting when the end of a 10 GiB gzip is needed!

See Installation for Ubuntu, the Release page for executables for your platform, and Compilation in case you want to compile the tool.

Considerations

  • Please, note that the initial complete index creation for an already gzip-compressed file (-i) still consumes as much time as a complete file decompression.
    Once created the index will reduce access time to the gzip file data.

Nonetheless, note that gztool creates index interleaved with extraction of data (-b), so in the practice there's no waste of time. Note that if extraction of data or just index creation are stopped at any moment, gztool will reuse the remaining index on the next run over the same data, so time consumption is always minimized.

Also gztool can monitor a growing gzip file (for example, a log created by rsyslog directly in gzip format) and generate the index on-the-fly, thus reducing in the practice to zero the time of index creation. See the -S (Supervise) option.

  • Index size is about 0.33% of a compressed gzip file size if the index is created after the file was compressed, or 10-100 times smaller (just a few Bytes or kiB) if gztool itself compresses the data (-c).

Note that the size of the index depends on the span between index points on the uncompressed stream - by default it is 10 MiB: this means that when retrieving randomly situated data only 10/2 = 5 MiB of uncompressed data must be decompressed (on average) no matter the size of the gzip file - which is a fairly low value!
The span between index points can be adjusted with -s (span) option (the minimum is -s 1 or 1 MiB).
For example, a span of -s 20 will create indexes half the size, and -s 5 will create indexes twice bigger.

Background

By default gzip-compressed files cannot be accessed in random mode: any byte required at position N requires the complete gzip file to be decompressed from the beginning to the N byte.
Nonetheless Mark Adler, the author of zlib, provided years ago a cryptic file named zran.c that creates an "index" of "windows" filled with 32 kiB of uncompressed data at different positions along the un/compressed file, which can be used to initialize the zlib library and make it behave as if compressed data begin there.

gztool builds upon zran.c to provide a useful command line tool.
Also, some optimizations and brand new behaviours have been added:

  • gztool can store line numbering information in the index using -[xX] (use only if source data is text!), and retrieve data from a specific line number using -L.
  • gztool can Supervise an still-growing gzip file (for example, a log created by rsyslog directly in gzip format) and generate the index on-the-fly, thus reducing in the practice to zero the time of index creation. See -S.
  • extraction of data and index creation are interleaved, so there's no waste of time for the index creation.
  • index files are reusable, so they can be stopped at any time and reused and/or completed later.
  • an ex novo index file format has been created to store the index
  • span between index points is raised by default from 1 MiB to 10 MiB, and can be adjusted with -s (span).
  • windows are compressed in file
  • windows are not loaded in memory unless they're needed, so the application memory footprint is fairly low (< 1 MiB)
  • gztool can compress files (-c) and at the same time generate an index that is about 10-100 times smaller than if the index is generated after the file has already been compressed with gzip.
  • Compatible with bgzip files (short-uncompressed complete-gzip-block sizes)
  • Compatible with complete gzip concatenated files (aka gzip members)
  • Compatible with rsyslog's veryRobustZip omfile option (variable-short-uncompressed complete-gzip-block sizes)
  • data can be provided from/to stdin/stdout
  • gztool can be used to remotely retrieve just a small part of a bigger gzip compressed file and sucessfully decompress it locally. See this stackexchange thread. Just the index must be also remotely available.

Installation

  • In Ubuntu, using my repository:

      sudo add-apt-repository ppa:roberto.s.galende/gztool
      sudo apt-get update
      sudo apt-get install gztool
    
  • See the Release page for executables for your platform. If none fit your needs, gztool is very easy to compile: see next sections.

Compilation

zlib.a archive library is needed in order to compile gztool: the package providing it is actually zlib1g-dev (this may vary on your system):

$ sudo apt-get install zlib1g-dev

$ gcc -O3 -o gztool gztool.c -lz -lm

Compilation in Windows

Compilation in Windows is possible using gcc for Windows and compiling the original zlib code to obtain the needed archive library libz.a.
Please, note that executables for different platforms are provided on the Release page.

  • download gcc for Windows: mingw-w64

  • Install it and add the path for gcc.exe to your Windows PATH

  • Download zlib code and compile it with your new gcc: zlib

  • The previous step generates the file zlib.a that you need to compile gztool: Copy gztool.c to the directory where you compiled zlib, and do:

    gcc -static -O3 -I. -o gztool gztool.c libz.a -lm

Usage

  gztool (v0.10.9)
  GZIP files indexer, compressor and data retriever.
  Create small indexes for gzipped files and use them
  for quick and random positioned data extraction.
  No more waiting when the end of a 10 GiB gzip is needed!
  //github.com/circulosmeos/gztool (by Roberto S. Galende)

  $ gztool [-[abLsv] #] [-cCdDeEfFhilStTwWxX|u[cCdD]] [-I <INDEX>] <FILE>...

  Note that actions `-bcStT` proceed to an index file creation (if
  none exists) INTERLEAVED with data flow. As data flow and
  index creation occur at the same time there's no waste of time.
  Also you can interrupt actions at any moment and the remaining
  index file will be reused (and completed if necessary) on the
  next gztool run over the same data.

 -a #: Await # seconds between reads when `-[ST]|Ec`. Default is 4 s.
 -b #: extract data from indicated uncompressed byte position of
     gzip file (creating or reusing an index file) to STDOUT.
     Accepts '0', '0x', and suffixes 'kmgtpe' (^10) 'KMGTPE' (^2).
 -C: always create a 'Complete' index file, ignoring possible errors.
 -c: compress a file like with gzip, creating an index at the same time.
 -d: decompress a file like with gzip.
 -D: do not delete original file when using `-[cd]`.
 -e: if multiple files are indicated, continue on error (if any).
 -E: end processing on first GZIP end of file marker at EOF.
     Nonetheless with `-c`, `-E` waits for more data even at EOF.
 -f: force file overwriting if destination file already exists.
 -F: force index creation/completion first, and then action: if
     `-F` is not used, index is created interleaved with actions.
 -h: print brief help; `-hh` prints this help.
 -i: create index for indicated gzip file (For 'file.gz' the default
     index file name will be 'file.gzi'). This is the default action.
 -I INDEX: index file name will be 'INDEX'.
 -l: check and list info contained in indicated index file.
     `-ll` and `-lll` increase the level of index checking detail.
 -L #: extract data from indicated uncompressed line position of
     gzip file (creating or reusing an index file) to STDOUT.
     Accepts '0', '0x', and suffixes 'kmgtpe' (^10) 'KMGTPE' (^2).
 -s #: span in uncompressed MiB between index points when
     creating the index. By default is `10`.
 -S: Supervise indicated file: create a growing index,
     for a still-growing gzip file. (`-i` is implicit).
 -t: tail (extract last bytes) to STDOUT on indicated gzip file.
 -T: tail (extract last bytes) to STDOUT on indicated still-growing
     gzip file, and continue Supervising & extracting to STDOUT.
 -u [cCdD]: utility to compress (`-u c`) or decompress (`-u d`)
          zlib-format files to STDOUT. Use `-u C` and `-u D`
          to manage raw compressed files. No index involved.
 -v #: output verbosity: from `0` (none) to `5` (nuts).
     Default is `1` (normal).
 -w: wait for creation if file doesn't exist, when using `-[cdST]`.
 -W: do not Write index to disk. But if one is already available
     read and use it. Useful if the index is still under a `-S` run.
 -x: create index with line number information (win/*nix compatible).
     (Index counts last line even w/o newline char (`wc` does not!)).
 -X: like `-x`, but newline character is '\r' (old mac).

  Example: Extract data from 1 GiB byte (byte 2^30) on,
  from `myfile.gz` to the file `myfile.txt`. Also gztool will
  create (or reuse, or complete) an index file named `myfile.gzi`:
  $ gztool -b 1G myfile.gz > myfile.txt

Please, note that STDOUT is used for data extraction with -bLtTu modifiers.

Examples of use

  • Make an index for test.gz. The index will be named test.gzi:

      $ gztool -i test.gz
    
  • Make an index for test.gz with name test.index:

      $ gztool -I test.index test.gz
    
  • Retrieve data from uncompressed byte position 1000000 inside test.gz. An index file will be created at the same time (named test.gzi):

      $ gztool -b 1m test.gz
    
  • Supervise an still-growing gzip file and generate the index for it on-the-fly. The index file name will be openldap.log.gzi in this case. gztool will execute until interrupted or until the supervised gzip file is overwritten from the beginning (this happens usually with compressed log files when they are rotated). It can also stop at first end-of-gzip data with -E.

      $ gztool -S openldap.log.gz
    
  • The previous command can be sent to background and with no verbosity, so we can forget about it:

      $ gztool -v0 -S openldap.log.gz &
    
  • Creating and index for all "*gz" files in a directory:

      $ gztool -i *gz
    
      ACTION: Create index
    
      Index file 'data.gzi' already exists and will be used.
      (Use `-f` to force overwriting.)
      Processing 'data.gz' ...
      Index already complete. Nothing to do.
    
      Processing 'data_project.2.tar.gz' ...
      Built index with 73 access points.
      Index written to 'data_project.2.tar.gzi'.
    
      Processing 'project_2.gz' ...
      Built index with 3 access points.
      Index written to 'project_2.gzi'.
    
  • Extract data from project.gz byte at 1 GiB to STDOUT, and use grep on this output. Index file name will be project.gzi:

      $ gztool -b 1G project.gz | grep -i "balance = "
    
  • Please, note that STDOUT is used for data extraction with -bcdtT modifiers, so an explicit command line redirection is needed if output is to be stored in a file:

      $ gztool -b 99m project.gz > uncompressed.data
    
  • Extract data from a gzipped file which index is still growing with a gztool -S process that is monitoring the (still-growing) gzip file: in this case the use of -W will not try to update the index on disk so that the other process is not disturb! (this is so because gztool always tries to update the index used if it thinks it's necessary):

      $ gztool -Wb 100k still-growing-gzip-file.gz > mytext
    
  • Extract data from line 10 million, to STDOUT:

      $ gztool -L 10m compressed_text_file.gz
    
  • Nonetheless note that if in the precedent example an index was previously created for the gzip file without the -x parameter (or not using -L), as it doesn't contain line numbering info, gztool will complain and stop. This can be circumvented by telling gztool to use another new index file name (-I), or even not using anyone at all with -W (do not write index) and an index file name that doesn't exists (in this case None - it won't be created because of -W), and so ((just) this time) the gzip will be processed from the beginning:

      $ gztool -L 10m -WI None compressed_text_file.gz
    
  • To tail to stdout, like a tail -f, an still-growing gzip file (an index file will be created with name still-growing-gzip-file.gzi in this case):

      $ gztool -T still-growing-gzip-file.gz
    
  • More on files still being "Supervised" (-S) by another gztool instance: they can also be tailed à la tail -f without updating the index on disk using -W:

      $ gztool -WT still-growing-gzip-file.gz
    
  • Compress (-c) an still growing (-E) file: in this case both still-growing-file.gz and still-growing-file.gzi files will be created on-the-fly as the source file grows. Note that in order to terminate compression, Ctrl+C must be used to kill gztool: this results in an incomplete-gzip-file as per GZIP standard, but this is not important as it will contain all the source data, and both gzip and gztool (or any other tool) can correctly and completely decompress it.

      $ gztool -Ec still-growing-file
    
  • If you have an incomplete index file (they just do not have the length of the source data, as it didn't correctly finish) and want to make it complete and so that the length of the uncompressed data be stored, just unconditionally complete it with -C with a new -i run over your gzip file: note that as the existent index data is used (in this case the file my-incomplete-gzip-data.gzi), only last compressed bytes are decompressed to complete this action:

      $ gztool -Ci my-incomplete-gzip-data.gz
    
  • Decompress a file like with gzip (-d), but do not delete (-D) the original one: Decompressed file will be myfile. Note that gzipped file must have a ".gz" extension or gztool will complain.

      $ gztool -Dd myfile.gz
    
  • Decompress a file that do not have ".gz" file extension, like with gzip (-d):

      $ cat mycompressedfile | gztool -d > my_uncompressed_file
    
  • Show internals of all index files in this directory. -e is used to not stop the process on the first error, if a *.gzi file is not a valid gzip index file. The -ll list option repetition will show data about each index point. -lll also decompress each point's window to ensure index integrity:

      $ gztool -ell *.gzi
    
          Checking index file 'accounting.gzi' ...
          Size of index file:        184577 Bytes (0.37%/gzip)
          Guessed gzip file name:    'accounting.gz' (66.05%) ( 50172261 Bytes )
          Number of index points:    15
          Size of uncompressed file: 147773440 Bytes
          List of points:
          @ compressed/uncompressed byte (index data size in Bytes @window's beginning at index file), ...
          #1: @ 10 / 0 ( 0 @56 ), #2: @ 3059779 / 10495261 ( 13127 @80 ), #3: @ 6418423 / 21210594 ( 6818 @13231 ), #4: @ 9534259 / 31720206 ( 7238 @20073 )...
      ...
    

If gztool finds the gzip file companion of the index file, some statistics are shown, like the index/gzip size ratio, or the ratio of compression of the gzip file.

Also, if the gzip is complete, the size of the uncompressed data is shown. This number is interesting if the gzip file is bigger than 4 GiB, in which case gunzip -l cannot correctly calculate it as it is limited to a 32 bit counter, or if the gzip file is in bgzip format, in which case gunzip -l would only show data about the first block (< 64 kiB).

  • Note that gztool -l tries to guess the companion gzip file of the index looking for a file with the same name, but without the i of the .gzi file name extension, or without the .gzi. But the gzip file name can also be directly indicated with this format:

      $ gztool -l -I index_filename gzip_filename
    

In this latter case only a pair of index+gzip filenames can be indicated with each use.

Index file format

Index files are created by default with extension '.gzi' appended to the original file name of the gzipped file:

filename.gz     ->     filename.gzi

If the original file doesn't have ".gz" extension, ".gzi" will be appended - for example:

filename.tgz     ->     filename.tgz.gzi

There's a special header to mark ".gzi" files as index files usable for this app:

+-----------------+-----------------+
|   0x0 64 bits   |    "gzipindx"   |     ~     16 bytes = 128 bits
+-----------------+-----------------+

This is "version 0" header, that is, it does not contain lines information. The header indicating that the index contains lines information is a "version 1" header, differing only in the capital "X" (each index registry point in this case contains an additional 64-bit counter to take lines into account). Next versions (if any) would use "gzipindx" string with lower and capital letters following a binary counting as if they were binary digits.

+-----------------+-----------------+
|   0x0 64 bits   |    "gzipindX"   |     version 1 header (index was created with `-[xX]` parameter)
+-----------------+-----------------+

Note that this header has been built so that this format will be "compatible" with index files generated for bgzip-compressed files. bgzip files are totally compatible with gzip: they've just been made so every 64 kiB of uncompressed data the zlib library is restart, so they are composed of independent gzipped blocks one after another. The bgzip command can create index files for bgzipped files in less time and with much less space than with this tool as they're already almost random-access-capable. The first 64-bit-long of bgzip files is the count of index pairs that are next, so with this 0x0 header gztool-index-files can be ignored by bgzip command and so bgzipped and gzipped files and their indexes could live in the same folder without collision.

All numbers are stored in big-endian byte order (platform independently). Big-endian numbers are easier to interpret than little-endian ones when inspecting them with an hex editor (like od for example).

Next, and almost one-to-one pass of struct access is serialize to the file. access->have and access->size are both written even though they'd always be equal. If the index file is generated with -S or -T on a still-growing gzip file (or somehow the index hasn't been completed because the gzip data was still incomplete), the values on disk for access->have and access->size will be respectively 0x0..0 and "number of actual index points written" (both uint64_t) to mark this fact. access->size MAY be UINT64_MAX to avoid the need to write this value as the number of index points are added to the file: as the index is incremental the number of points can be determined by reading the index until EOF. access->have MAY also be greater than zero but lower than access->size: this can occur when an already finished index is increased with new points (source gzip may have grown) - in this case this is also considered an incomplete index: when the index be correctly closed both numbers will have the same value (a Ctrl+C before would leave the index "incomplete", but usable for next runs in which it can be finished).

After that, comes all the struct point data. As previously said, windows are compressed so a previous register (32 bits) with their length is needed. Note that an index point with a window of size zero is possible.

After all the struct point structures data, the original uncompressed data size of the gzipped file is stored (64 bits).

Please note that not all stored numbers are 64-bit long. This is because some counters will always fit in less length. Refer to code.

With 64 bit long numbers, the index could potentially manage files up to 2^64 = 16 EiB (16 777 216 TiB).

Regarding line number counting (-[xX]), note that gztool's index counts last line in uncompressed data even if the last char isn't a newline char - whilst wc command will not count it in this case!. Nonetheless, line counting when extracting data with -[bLtT] does follow wc convention - this is in order to not obtain different (+/-1) results reading gztool output info and wc counts.

Other tools which try to provide random access to gzipped files

  • also based on zlib's zran.c:

    • Perl module Gzip::RandomAccess. It seems to create the index only in memory, after decompressing the whole gzip file.
    • Go package gzran. It seems to create the index only in memory, after decompressing the whole gzip file.
    • jzran. It's a Java library based on the zran.c sample from zlib.
  • bgzip command, available in linux with package tabix (used for chromosome handling). This discussion about the implementation is very interesting: random-access-to-zlib-compressed-files. I've developed also a bgztail command tool to tail bgzipped files, even as they grow.

  • dictzip command, is a format compatible with gzip that stores an index in the header of the file. Uncompressed size is limited to 4 GiB - see also idzip below. The dictionary header cannot be expanded if more gzip data is added, and it cannot be added to an existent gzip file - both issues are succesfully managed by gztool.

  • idzip Python command and function, builds upon dictzip to overcome the 4 GiB limit of dictzip by using multiple gzip members.

  • GZinga: Seekable and Splittable GZip, provides Java language classes to create gzip-compatible compressed files using the Z_FULL_FLUSH option, to later access or split them.

  • indexed_gzip Fast random access of gzip files in Python: it also creates file indexes, but they are not as small and they cannot be reused as easily as with gztool.

  • lzopfs. Random access to compressed files with a FUSE filesystem: allows random access to lzop, gzip, bzip2 and xz files - lzopfs allows to mount gzip, bzip2, lzo, and lzma compressed files for random read-only access. I.e., files can be seeked arbitrarily without a performance penalty. It hasn't been updated for years, but there's a fork that provides CMake installation. See also this description of its internals.

  • pymzML v2.0. One key feature of pymzML is the ability to write and read indexed gzip files. It has its own index format. Read also this.

  • zindex creates SQLite indexes for text files based on regular expressions.

Other interesting links

Version

This version is v0.10.9.

Please, read the Disclaimer. This is still a beta release. In case of any errors, please open an issue.

License

A work by Roberto S. Galende
distributed under the same License terms covering
zlib from Mark Adler (aka Zlib license):
  This software is provided 'as-is', without any express or implied
  warranty.  In no event will the authors be held liable for any damages
  arising from the use of this software.
  Permission is granted to anyone to use this software for any purpose,
  including commercial applications, and to alter it and redistribute it
  freely, subject to the following restrictions:
  1. The origin of this software must not be misrepresented; you must not
     claim that you wrote the original software. If you use this software
     in a product, an acknowledgment in the product documentation would be
     appreciated but is not required.
  2. Altered source versions must be plainly marked as such, and must not be
     misrepresented as being the original software.
  3. This notice may not be removed or altered from any source distribution.

/* zlib.h -- interface of the 'zlib' general purpose compression library
  version 1.2.11, January 15th, 2017
  Copyright (C) 1995-2017 Jean-loup Gailly and Mark Adler
  This software is provided 'as-is', without any express or implied
  warranty.  In no event will the authors be held liable for any damages
  arising from the use of this software.
  Permission is granted to anyone to use this software for any purpose,
  including commercial applications, and to alter it and redistribute it
  freely, subject to the following restrictions:
  1. The origin of this software must not be misrepresented; you must not
     claim that you wrote the original software. If you use this software
     in a product, an acknowledgment in the product documentation would be
     appreciated but is not required.
  2. Altered source versions must be plainly marked as such, and must not be
     misrepresented as being the original software.
  3. This notice may not be removed or altered from any source distribution.
  Jean-loup Gailly        Mark Adler
  jloup@gzip.org          madler@alumni.caltech.edu
  The data format used by the zlib library is described by RFCs (Request for
  Comments) 1950 to 1952 in the files http://tools.ietf.org/html/rfc1950
  (zlib format), rfc1951 (deflate format) and rfc1952 (gzip format).
*/

Disclaimer

This software is provided "as is", without warranty of any kind, express or implied. In no event will the authors be held liable for any damages arising from the use of this software.

Author

by Roberto S. Galende

on code by Mark Adler's zlib.

You can’t perform that action at this time.