moc-set

A command-line tool to build, update and query a persistent set of Multi-Order Coverages maps (MOCs), see this IVOA standard.

About

Query a set of S-MOCs pre-saved in a single large binary file.
The file can be seen as a persistent cache preventing from having to open/read/parse a possible large set of FITS files.
A moc-set can be updated by adding or removing MOCs from it, without having to re-build the full binary file (except from time to time using the purge command). It is built in such a way that read operations are allowed while updating (add, remove, change status, purge) the binary file, but a write operation immediately fails (return status != 0) if another write operation is already ongoing.

Used by

MOC set in used:

• in production in VizieR when querying a cone on the > 16_000 catalogues
• in an ongoing project leading by @lmichel to look for XMM FOVs intersecting a user define region.

Similar to the CDS MOC Server.

The CDS MOC Server additionally stores metadata information associated to each MOC, allows queries with constraints on those metadata, and supports T-MOCs and ST-MOCs in addition to S-MOCs.

So, moc-set offers functionalities similar to a sub-part of the MOC Server functionalities (except for T-MOCs and ST-MOCs, one could build a MOC Server like service based on moc-set, using an additional mechanism to store and filter metadata).

The MOC Server is implemented in Java. All information used by the MOC Server are loaded in memory at service startup. To speed up the starting time, the MOC Server also resort on a binary dump, needing synchronisation mechanisms with the running server data when the list of MOCs is updated.

moc-set has primarily been developed for VizieR needs: lighter/simpler version of the MOC server, with less information needed (VizieR already store catalogues metadata), system programming language with no runtime, possibly less memory available, ...

Install

From pypi for python users

MOCSet is available in pypi, you can thus install the mocset executable using pip:

pip install -U moc-set
mocset --help

Debian package

Download the last mocset_vxx_yyy.deb corresponding to your architecture (x86_64_musl has the most chances to fit your needs) from the github release page.

Install the .deb by clicking on it or using the command line:

sudo dpkg -i mocset_vxx_yyy.deb
sudo apt-get install -f

Then you can use the tool:

mocset
man mocset

You can uninstall using, e.g.:

sudo dpkg -r $(dpkg -f mocset_vxx_yyy.deb Package)

Pre-compile binaries for MacOS, Linux and Windows

Download the last mocset-vxx_yyy.tar.gz corresponding to your architecture from the github release page. You probably want ot use:

• Linux: mocset-vxx-x86_64-unknown-linux-musl.tar.gz
• MacOS: mocset-vxx-x86_64-apple-darwin.tar.gz
• Windows: mocset-vxx-.zip

WARNING: for linux, use musl instead of gnu (high chances of incompatibility in the latter case)

The tar contains a single executable binary file.

tar xzvf mocset-vxx-yyy.tar.gz
./mocset

Compile from source code

Install rust (and check that ~/.cargo/bin/ is in your path), or update the Rust compiler with:

rustup update

Clone the moc lib rust project:

git clone https://github.com/cds-astro/cds-moc-rust

Install from using cargo:

cargo install --path crates/set

Usage

Exec mocset --help to get the help message:

> mocset --help
...
USAGE:
    mocset <SUBCOMMAND>

OPTIONS:
    -h, --help       Print help information
    -V, --version    Print version information

SUBCOMMANDS:
    append       Append the given MOCs to an existing mocset
    chgstatus    Change the status flag of the given MOCs identifiers (valid, deprecated,
                     removed)
    extract      Extracts a MOC from the given moc-set
    help         Print this message or the help of the given subcommand(s)
    list         Provide the list of the MOCs in the mocset and the associated flags
    make         Make a new mocset
    purge        Purge the mocset removing physically the MOCs flagged as 'removed'
    query        Query the mocset

MOC set creation

Simply list in a csv file a set of identifier (integer in [0, 281_474_976_710_655]), and MOC FITS file path. See e.g the file moclist.txt:

1,data/CDS_B_bax_bax.fits
2,data/CDS_I_172_irs.fits
3,data/CDS_I_274_ccdm.fits
4,data/CDS_I_42_catalog.fits
5,data/CDS_II_203A_faint.fits
6,data/CDS_II_261_GOODS_S.fits
7,data/CDS_II_339_summary.fits
8,data/CDS_II_97_ans.fits
9,data/CDS_III_206_orion.fits
10,data/CDS_III_43_main.fits
...

Then execute moc-set from the mocset command:

> mocset make --n128 3 --moc-list moclist.txt --delimiter , mocset.bin

or from the source code, installing the Rust compiler (including Cargo):

> cargo run --release -- make --n128 3 --moc-list moclist.txt --delimiter , mocset.bin

To build the moc-set binary file mocset.bin.

The total number of MOCs that can be stored in the file equals `n128 * 128 - 1 (so 383 in the example above, since n128=3).

See the help message for more details:

> mocset make --help
...
Make a new mocset

USAGE:
    mocset make [OPTIONS] <OUTPUT>

ARGS:
    <OUTPUT>    Output file, storing the MOC set

OPTIONS:
    -d, --delimiter <SEPARATOR>       Delimiter used to separate the moc identifier from the moc
                                      path [default: " "]
    -h, --help                        Print help information
    -l, --moc-list <MOC_LIST_PATH>    Input file containing the 'moc_id moc_path' list (default:
                                      read from stdin) 'moc_id' must be a positive integer smaller
                                      than 281_474_976_710_655 (can be stored on 6 bytes). Use a
                                      negative value to flag as deprecated
    -n, --n128 <N128>                 n x 128 - 1 = number of MOCs that can be stored in this moc
                                      set [default: 1]

List the content of a MOC set

Simply use mocset list:

> mocset list mocset.bin
id,status,depth,n_ranges,byte_size
1,valid,11,1673,13384
2,valid,11,36022,288176
3,valid,11,55470,443760
4,valid,11,587,4696
5,valid,11,534,4272
6,valid,11,17,136
...

• status: can be valid, deprecated or removed
• n_ranges: number of ranges in the MOC
• byte_size: size of the MOC stored in the MOC set (in bytes)

See the help message:

> mocset list --help
... 
Provide the list of the MOCs in the mocset and the associated flags

USAGE:
    mocset list [OPTIONS] <FILE>

ARGS:
    <FILE>    The moc-set to be read

OPTIONS:
    -h, --help      Print help information
    -r, --ranges    Print byte ranges instead of byte_size

Querying a MOC set

See the help to get the different query modes:

> mocset query --help
...
USAGE:
    mocset query [OPTIONS] <FILE> <SUBCOMMAND>

ARGS:
    <FILE>    The moc-set to be read

OPTIONS:
    -c, --print-coverage    Print in output the sky fraction (in '[0.0, 1.0]') covered by each
                            selected MOC
    -d, --add-deprecated    Also selects MOCs flagged as deprecated
    -h, --help              Print help information

SUBCOMMANDS:
    cone    A cone, i.e. a position with a small area around (approximated by a MOC)
    help    Print this message or the help of the given subcommand(s)
    moc     The given MOC (you create a moc using moc-cli and pipe it into moc-set)
    pos     Single position

Query for a single position

To obtain the list of the MOCs containing a the position 90.0 +0.0 (in decimal degrees), use:

> mocset query moclist.bin pos 90.0 +0.0
id
161

The result is the list of the identifier of the MOC covering the input position (only one MOC, with identifier 161 in our example).

See the full help:

> mocset query pos --help
...
Single position

USAGE:
    mocset query <FILE> pos <LON_DEG> <LAT_DEG>

ARGS:
    <LON_DEG>    Longitude of the cone center (in degrees)
    <LAT_DEG>    Latitude of the cone center (in degrees)

OPTIONS:
    -h, --help    Print help information

Query for a cone

To obtain the list of the MOCs overlapping the cone of center 90.0 +0.0 and having a radius of 320 arcsec, use:

> mocset query mocset.bin cone 90.0 +0.0 320.0 --precision 5
id
2
3
35
161

--precision is optional, its default value is 2:

• at a precision of 0, mocset transforms the input cone in a MOC containing 1 to 9 values.
• at a precision of 1, mocset transforms the input cone in a MOC containing 1 to 9x4 values.
• at a precision of n, mocset transforms the input cone in a MOC containing 1 to 9x4^n values.

See the help:

> mocset query cone --help
...
A cone, i.e. a position with a small area around (approximated by a MOC)

USAGE:
    mocset query <FILE> cone [OPTIONS] <LON_DEG> <LAT_DEG> <R_ARCSEC>

ARGS:
    <LON_DEG>     Longitude of the cone center (in degrees)
    <LAT_DEG>     Latitude of the cone center (in degrees)
    <R_ARCSEC>    Radius of the cone (in arcseconds)

OPTIONS:
    -h, --help                Print help information
    -i, --included            Returns MOCs containing the whole cone MOC (instead of overlapping
                              only)
    -p, --precision <PREC>    MOC precision; 0: depth 'd' at which the cone is overlapped by 1 to
                              max 9 cells; 1: depth 'd' + 1; n: depth 'd' + n [default: 2]

Query for a MOC

To obtain the list of the MOCs overlapping the MOC stored in a given FITs file, e.g the MOC of XMM4 DR9, use:

> mocset query mocset.bin moc YOUR_PATH/CDS_IX_59_xmm4dr9s.fits
id
1
2
3
4
5
6
7
8
9
10
...

You may also use a MOC generate on the fly using moc-cli

> moc from polygon 5 "(0.0,0.0),(10.0,0.0),(0.0,10.0)" ascii | mocset query mocset.bin moc - --format ascii
id
1
2
3
4
7
8
13
26
27
35
36
60
63
...

See the help

> mocset query moc --help
...
The given MOC (you create a moc using moc-cli and pipe it into moc-set)

USAGE:
    mocset query <FILE> moc [OPTIONS] <INPUT>

ARGS:
    <INPUT>    Path of the input MOC file (or stdin if equals "-")

OPTIONS:
    -f, --format <INPUT_FMT>    Format of the input MOC ('ascii', 'json' or 'fits') [default: guess
                                from the file extension]
    -h, --help                  Print help information
    -i, --included              Returns MOCs containing the whole given MOC (instead of overlapping)

Generic parameter

For all queries, the option --add-deprecated (or -d) add the MOCs flagged as deprecated in the output, e.g.:

mocset query --add-deprecated moclist.bin pos 90.0 +0.0 --precision 5

Updating a MOC set

You can:

• chgstatus change the status of a MOC in the MOC set to flag it as deprecated or back to valid, or as removed (WARNING: once removed, a MOC cannot be flagged back as deprecated or valid)
• append a new MOC at the end of the MOC set, the ID of a removed MOC can be reused
• purge a MOC set to physically remove all MOCs flagged as removed. This process basically re-write the whole MOC set file.

Example of usage of chgstatus

> mocset list mocset.bin
id,status,depth,n_ranges,byte_size
1,valid,11,1673,13384
2,valid,11,36022,288176
...
> mocset chgstatus mocset.bin 1 deprecated
> mocset list mocset.bin
id,status,depth,n_ranges,byte_size
1,deprecated,11,1673,13384
2,valid,11,36022,288176
...
> mocset chgstatus mocset.bin 1 valid
> mocset list mocset.bin
id,status,depth,n_ranges,byte_size
1,valid,11,1673,13384
2,valid,11,36022,288176
...
> mocset chgstatus mocset.bin 1 removed
> mocset list mocset.bin
id,status,depth,n_ranges,byte_size
1,removed,11,1673,13384
2,valid,11,36022,288176
...

Example of usage of append:

> mocset append mocset.bin 3000 MY_PATH/my_new_moc.fits

Example of usage of purge:

> mocset purge mocset.bin

Help messages:

> mocset chgstatus --help
... 
Change the status flag of the given MOCs identifiers (valid, deprecated, removed)

USAGE:
    mocset chgstatus <FILE> <ID> <NEW_STATUS>

ARGS:
    <FILE>          The moc-set to be updated
    <ID>            Identifier of the MOC we want to modify the flag
    <NEW_STATUS>    New status flag to be set

OPTIONS:
    -h, --help    Print help information

> mocset append --help
...
Append the given MOCs to an existing mocset

USAGE:
    mocset append <FILE> <ID> <MOC_PATH>

ARGS:
    <FILE>        The moc-set to be updated
    <ID>          Identifier of the MOC we want to append. 'moc_id' must be a positive integer
                  smaller than 281_474_976_710_655 (can be stored on 6 bytes). Use a negative
                  value to flag as deprecated
    <MOC_PATH>    Path of the MOC to be append

OPTIONS:
    -h, --help    Print help information

> mocset purge --help
...
Purge the mocset removing physically the MOCs flagged as 'removed'

USAGE:
    mocset purge [OPTIONS] <FILE>

ARGS:
    <FILE>    The moc-set file to be purge

OPTIONS:
    -h, --help           Print help information
    -n, --n128 <N128>    n x 128 - 1 = number of MOCs that can be stored in this purged moc set, if
                         larger than the current value

Implementation information

Principle

We resort on mapped files and the system disk cache. Although the first moc-set query may still be quite long, the performances of subsequent queries will depend on the ability of the system to keep the file's data in the disk cache.

Remark

If used on a server with a lot of I/O, the data may not stay in the disk cache for long. If we nevertheless want to always answer quickly, there is at least three options:

• write a daemon loading the full dataset in memory
• create a VM (or a docker?) containing only the moc-set tool
• my favourite one: use shared memory, resorting e.g. on the shared_memory crate

File structure

The structure of the file is made of two main blocks:

• a fixed length header, made of
  • a fixed length metadata
  • a fixed length cumulative byte index
• a extensible data part

Metadata part

A fixed size (chosen at creation) buffer of n u64 elements.

• the first u64 contains an indication on the buffer size
• all other u64 values, one per MOC in the moc-set, are made of 3 components:
  • a u8 flag taking 4 possible values:
    • Valid MOC
    • Deprecated MOC
    • Removed MOC
    • Free slot
  • the depth of the MOC, stored on `u8
  • the MOC identifier, stored on the 6 remaining bytes.

We stop reading the metadata when we encounter the first Free flag.

The file can store n - 1 MOCs.

Cumulative byte index

A fixed size (chosen at creation) buffer of n u64 elements.

• the first u64 contains the byte size of the header, i.e. the size of the file with no data it also correspond to the starting byte of the first MOC that will be inserted in the moc-set;
• all other elements at index i contains the size of the file if only storing the $i+1$ first MOCs.

To get the size (in bytes) of the MOC at index i, we can simply perform the operation: cumulIndex[i + 1] - cumulIndex[i].

Data part

A simple concatenation of the MOCs ranges.
According to the depth of each MOC, the ranges are either ranges of u32 or ranges of u64.

Updates and Reading while writing

The only MOC information that can be modified (re-written) is its status flag.
But the status flag is coded on a single byte and a CPU read and write single bytes atomically.
Although, several cores may see the byte in a different state, it is not critical: a MOC flagged as 'deprecated' or 'deleted' is still physically stored in the moc-set, and we can accept a small delay in changing its state.

• Read operation order
  • do not need a read lock
  • read the metadata part of the header
  • read the cumulative bytes index in the header
  • read the MOC(s)
• Write Operation order
  • ask for a write lock (create a .lock file)
  • append the new MOC(s) to the end of the file
  • append the new entries to the cumulative bytes index
  • possibly update the flag of existing MOCs and append new entries to the end of the header

The header, i.e. both the metadata and the cumulative byte index, is of fixed size, with free space reserved at file creation.
If a problem occurs while writing, the update is simply going to be ignored (except the part written in the meta): at the next write operation, if the meta has not been written (and it is the last part to be written), the code assumes that the cumulative index beyond the know meta elements is free, the same for the data byte beyond the last known cumulative index byte.

If a process is reading while another is writing, the newly written part will also simply be ignored if the metadata part has not yet been updated/written.

It works well because we only append new MOCs (we never delete information), we only update metadata to flag a MOC as removed of deprecated, and we respect the read/write order as described above.

Purge

The purge operation is the only one removing physically stored MOCs.
It actually consists in building a new moc-set, removing the MOCs flagged as removed, in a temporary file, and then making a move replacing the previous moc-set by the new one.
The purge operation uses a write lock.

ToDo list

• Create compact MOCs in which ranges of a single element (at the moc max_order) are replaced by a single value with a flag on the 2 LSB telling if the value id
  • 0: a single cell
  • 1: a range starting value
  • 2: a range ending value

Recall

• Drop linux disk cache, in root mode: sync; echo 3 > /proc/sys/vm/drop_caches
• Build for being compatible with a large range of systems, avoiding the GLIBC version error (see this post):

cargo build --target x86_64-unknown-linux-musl