NAME

CHI::Cascade - a cache dependencies (cache and like 'make' utility concept)

SYNOPSIS

    use CHI;
    use CHI::Cascade;

    $cascade = CHI::Cascade->new(chi => CHI->new(...));

    $cascade->rule(
        target  => 'unique_name',
        depends => ['unique_name_other1', 'unique_name_other2'],
        code    => sub {
            my ($rule, $target_name, $values_of_depends) = @_;

            # $values_of_depends == {
            #     unique_name_other1 => $value_1,
            #     unique_name_other2 => $value_2
            # }
            # $rule->target     eq      $target_name
            # $rule->depends    ===     ['unique_name_other1', 'unique_name_other2']
            # $rule->dep_values ==      $values_of_depends
            # $rule->params     ==      { a => 1, b => 2 }

            # Now we can calcualte $value
            return $value;
        },
        params  => { a => 1, b => 2 }
    );

    $cascade->rule(
        target  => 'unique_name_other1',
        depends => 'unique_name_other3',
        code    => sub {
            my ($rule, $target_name, $values_of_depends) = @_;

            # $values_of_depends == {
            #     unique_name_other3 => $value_3
            # }

            # computing here
            return $value;
        }
    );

    $value_of_this_target = $cascade->run('unique_name');

DESCRIPTION

This module is the attempt to use a benefits of caching and 'make' concept. If we have many an expensive tasks (a computations or sometimes here used term as a recomputing) and want to cache it we can split its to small expsnsive tasks and to describe dependencies for cache items.

This module is experimental yet. I plan to improve it near time but some things already work. You can take a look for t/* tests as examples.

CONSTRUCTOR

$cascade = CHI::Cascade->new( %options )

This method constructs a new CHI::Cascade object and returns it. Key/value pair arguments may be provided to set up the initial state. Options are:

chi

Required. Instance of CHI object. The CHI::Cascade doesn't construct this object for you. Please create instance of CHI yourself.

busy_lock

Optional. Default is never. This is not busy_lock option of CHI! This is amount of time (to see "DURATION EXPRESSIONS" in CHI) until all target locks expire. When a target is to being computing it is locked. If process which is to be computing target and it will die or OS will be hangs up we can dead locks and locked target will never recomputed again. This option helps to avoid it. You can set up a special busy_lock for rules too.

target_chi

Optional. This is CHI cache for target markers. Default value is value of "chi" option. It can be useful if you use a "l1_cache" in CHI option. So you can separate data of targets from target markers - data will be kept in a file cache and a marker in memory cache for example.

METHODS

rule( %options )

To add new rule to CHI::Cascade object. All rules should be added before first "run" method

The keys of %options are (options are passed directly in CHI::Cascade::Rule constructor):

target

Required. A target for "run" and for searching of "depends". It can be as scalar text or Regexp object created through qr//

depends

Optional. The scalar, arrayref or coderef values of dependencies. This is the definition of target(s) from which this current rule is dependent. If depends is:

scalar

It should be plain text of single dependence of this target.

arrayref

An each item of list can be scalar value (exactly matched target) or code reference. If item is coderef it will be executed once as $coderef->( $rule, $rule->qr_params ) and should return a scalar value as current dependence for this target at runtime (the API for coderef parameters was changed since v0.16)

coderef

This subroutine will be executed once inside run method if necessary with parameters as: $coderef->( $rule, <$rule-qr_params|CHI::Cascade::Rule/qr_params >> ) (API was changed since v0.16). It should return scalar or arrayref. The returned value is scalar it will be considered as single dependence of this target and the behavior will be exactly as described for scalar in this paragraph. If the returned value is arrayref it will be considered as list of dependencies for this target and the behavior will be exactly as described for arrayref in this paragraph.

depends_catch

Optional. This is coderef for dependence exceptions. If any dependence from list of "depends"'s option throws an exception of type CHI::Cascade::Value by die (for example like this code: die CHI::Cascade::Value->new->value( { i_have_problem => 1 } ) ) then the $cascade will execute this code as $rule->{depends_catch}->( $this_rule_obj, $exception_of_dependence, $rule_obj_of_dependence, $plain_text_target_of_dependence ) and you can do into inside a following:

re-die new exception of any type

If your new exception will be type of CHI::Cascade::Value you will get the value of this object from "run" method immediately (please to see "code" below) without saving in cache.

If exception will be other type this will be propogated onward beyond the "run" method

to do something

You can make something in this code. After execution of your code the cascade re-throws original exception of dependence like described above in "re-die" section.

But please notice that original exception has a status of "thrown from code" so it can be catched later by other "depends_catch" callback from other rule located closer to the call hierarchy of "run".

Please notice that there no way to continue a "code" of current rule if any dependence throws an exception!. It because that the main concept of execution code of rules is to have all valid values (cached or recomputed) of all dependencies before execution of dependent code.

code

Required. The code reference for computing a value of this target (a computational code). Will be executed if no value in cache for this target or any dependence or dependences of dependences and so on will be recomputed. Will be executed as $code->( $rule, $target, $hashref_to_value_of_dependencies ) (The API of running this code was changed since v0.10)

If you want to terminate a code and to return immediately from "run" method and don't want to save a value in cache you can throw an exception from "code" of type CHI::Cascade::Value. Your instance of CHI::Cascade::Value can have a value or cannot (a valid value can be even undef!). A "run" method returns either a value is set by you (through "value" in CHI::Cascade::Value method) or value from cache or undef in other cases. Please to see CHI::Cascade::Value

If "run" method will have a "defer" option as true this code will not be executed and you will get a set bit CASCADE_DEFERRED in "state" bit mask variable. This may useful when you want to control a target execution.

$rule

An instance of CHI::Cascade::Rule object. You can use it object as accessor for some current executed target data (plain text of target, for getting of parameters and so on). Please to see CHI::Cascade::Rule

$target

The current executed target as plain text for this "code"

$hashref_to_value_of_dependencies

A hash reference of values (values are cleaned values not CHI::Cascade::Value objects!) of all dependencies for current target. Keys in this hash are flat strings of dependecies and values are computed or cached ones.

This module should guarantee that values of dependencies will be valid values even if value is undef. This code can return undef value as a valid code return but author doesn't recommend it. If CHI::Cascade could not get a valid values of all dependencies of current target before execution of this code the last will not be executed (The run will return undef).

params

Optional. You can pass in your code any additional parameters by this option. These parameters are accessed in your rule's code through "params" in CHI::Cascade::Rule method of CHI::Cascade::Rule instance object.

busy_lock

Optional. Default is "busy_lock" of constructor or never if first is not defined. This is not busy_lock option of CHI! This is amount of time (to see "DURATION EXPRESSIONS" in CHI) until target lock expires. When a target is to being computed it is locked. If process which to be recomputing a target and it will die or OS will be hangs up we can dead locks and locked target will never recomputed again. This option helps to avoid it.

recomputed

Optional. This is a computational callback (coderef). If target of this rule was recomputed this callback will be executed right away after a recomputed value has been saved in cache. The callback will be executed as $coderef->( $rule, $target, $value ) where passed parameters are:

$rule

An instance of CHI::Cascade::Rule class. This instance is recreated for every target searching and recomputing if need.

$target

A current target as string

$value

The instance of CHI::Cascade::Value class. You can use a computed value as $value->value

For example you can use this callback for notifying of other sites that your target's value has been changed and is already in cache.

value_expires

Optional. Sets an CHI's cache expire value for all future target markers are created by this rule in notation described in "DURATION EXPRESSIONS" in CHI. The default is 'never'. It can be coderef or string scalar format as "DURATION EXPRESSIONS" in CHI. A coderef should return value in same format.

ttl

Optional. An arrayref for min & max intervals of TTL. Example: [ 60, 3600 ] - where the minimum ttl is seconds and the maximum is 3600 seconds. Targets of this rule will be recomputed during from 60 up to 3600 seconds from touched time of any dependence this rule. Please read "CASCADE_TTL_INVOLVED" in CHI::Cascade::Value too.

run( $target, %options )

This method makes a cascade computation if need and returns value (value is cleaned value not CHI::Cascade::Value object!) for this target If any dependence of this target of any dependencies of dependencies were (re)computed this target will be (re)computed too.

The run method of instance of cascade can be called from other run method of same instance and from callref function inside depends rule's option. This was made possible by creating a separate data instance for each root call of run method. This can come in handy when you compute dependencies on the go, which are computed by the same object (instance) of cascade.

$target

Required. Plain text string of target.

%options

Optional. And all options are optional too A hash of options. Valid keys and values are:

state

A scalarref of variable where will be stored a state of "run". Value will be a bit mask.

defer

If value will be a true then computational code will not be run if there is a need. After "run" you can test status of returned value - it should be (re)computed or not by bit CASCADE_DEFERRED in saved "state" variable. If the CASCADE_DEFERRED bit is set you can recall "run" method again or re-execute target in other process for a non-blocking execution of current process.

actual_term

The value in seconds (a floating point value) of actual term. The actual term is period when dependencies to be checked for $target in "run" method. If this option is not defined then the "run" method checks a dependencies of $target every time in runtime. But sometimes (when a target has many dependencies) we could want to reduce an amount of dependencies checks. For example if actual_term will be defined as 2.5 this will mean to check a dependencies only every 2.5 seconds. So recomputing in this example can be recomputed only one time in every 2.5 seconds (even if one from dependencies will be updated). But if value of $target is missing in cache a recomputing can be run regardless of this option.

ttl

A scalarref for getting current TTL for value of 'run' target. The TTL is "time to live" as TTL in DNS. If any rule in a path of following to dependencies has ttl parameter then the cascade will do there:

  1. will look up a time of this retouched dependence;

  2. if rule's target marker already has a upper time and this time in future the target will be recomputed in this time in future and before this moment you will get a old data from cache for 'run' target. If this time is there and has elapsed cascade will use a standard algorithm.

  3. will look up the rule's ttl parameter (min & max ttl values) and will generate upper time of computation of this rule's target and will return from "run" method old data of 'run' target. Next "run"s executions will return old values of any targets where this TTL-marked target is as dependence.

  4. In any case if old value misses in cache the cascade will recompute codes.

This feature was made for reset situation. For example if we have 'reset' rule and all rules depend from this one rule the better way will be to have 'ttl' parameter in every rule except 'reset' rule. So if rule 'reset' will be retouched (or deleted) other targets will be recomputed during time from 'min' and 'max' intervals from 'reset' touched time. It reduce a server's load. Later i will add examples for this and will document this feature more details. Please read "CASCADE_TTL_INVOLVED" in CHI::Cascade::Value too.

stash

A hashref to stash - temporary data container between rule's codes. Please see "stash ()" method for details.

touch( $target )

This method refreshes the time of this target. Here is analogy with touch utility of Unix and behaviour as make(1) after it. After "touch" all targets are dependent from this target will be recomputed at next "run" with an appropriate ones.

target_remove ( $target )

It's like a removing of target file in make. You can force to recompute target by this method. It will remove target marker if one exists and once when cascade will need target value it will be recomputed. In a during recomputing of course cascade will return an old value if one exists in cache.

stash()

Deprecated! It returns hashref to a stash. A stash is hash for temporary data between rule's codes. It can be used only from inside "run". Example:

    $cascade->run( 'target', stash => { key1 => value1 } )

and into rule's code:

    # DEPRECATED - OLD METHOD! It's supported and works but please don't use it
    $rule->cascade->stash->{key1}

    # NEW METHOD:
    $rule->stash->{key1}

If a "run" method didn't get stash hashref the default stash will be as empty hash. You can pass a data between rule's codes but it's recommended only in special cases. For example when run's target cannot get a full data from its target's name.

STATUS

This module is experimental and not finished for new features ;-) Please send me issues through https://github.com/Perlover/CHI-Cascade page

ANALOGIES WITH make

Here simple example how it works. Here is a direct analogy to Unix make utility:

    In CHI::Cascade:            In make:

    rule                        rule
    depends                     prerequisites
    code                        commands
    run( rule_name )            make target_name

FEATURES

The features of this module are following:

Computing inside process

If module needs to compute item for cache we compute inside process (no forks) For web applications it means that one process for one request could take a some time for computing. But other processes will not wait and will get either old previous computed value or undef value.

Non-blocking computing for concurrent processes

If other process want to get data from cache we should not block it. So concurrent process can get an old data if new computing is run or can get undef value. A concurrent process should decide itself what it should do after it - try again after few time or print some message like 'Please wait and try again' to user.

Each target is splitted is two items in cache

For optimization this module keeps target's info by separately from value item. A target item has lock & timestamp fields. A value item has a computed value.

EXAMPLE

For example please to see the SYNOPSIS

When we prepared a rules and a depends we can:

If unique_name_other1 and/or unique_name_other2 are(is) more newer than unique_name the unique_name will be recomputed. If in this example unique_name_other1 and unique_name_other2 are older than unique_name but the unique_name_other3 is newer than unique_name_other1 then unique_name_other1 will be recomputed and after the unique_name will be recomputed.

And even we can have a same rule:

    $cascade->rule(
        target  => qr/^unique_name_(.*)$/,
        depends => sub { 'unique_name_other_' . $_[1] },
        code    => sub {
            my ($rule, $target_name, $values_of_depends) = @_;

            # $rule->qr_params          === ( 3 )
            # $target_name              == 'unique_name_3' if $cascade->run('unique_name_3') was
            # $values_of_depends        == {
            #     unique_name_other_3   => $value_ref_3
            # }
        }
    );

    $cascade->rule(
        target  => qr/unique_name_other_(.*)/,
        code    => sub {
            my ($rule, $target_name, $values_of_depends) = @_;
            ...
        }
    );

When we will do:

    $cascade->run('unique_name_52');

$cascade will find rule with qr/^unique_name_(.*)$/, will make =~ and will find a depend as unique_name_other_52

AUTHOR

This module has been written by Perlover <perlover@perlover.com>

LICENSE

This module is free software and is published under the same terms as Perl itself.

SEE ALSO

CHI::Cascade::Rule

An instance of this object can be used in your target codes.

CHI

This object is used for cache.

CHI::Driver::Memcached::Fast

Recommended if you have the Memcached

CHI::Driver::File

Recommended if you want to use the file caching instead the Memcached for example