Persistent arrays

This page describes several implementations for in-memory Tcl arrays that are synchronized with persistent storage. They should not be confused with persistent data structures.

Plain text files

Richard Suchenwirth 2002-05-30 - Large data (e.g. a simple database) are in Tcl often stored in arrays. You can load array contents from files, and you can store them to files, so the data persists between sessions. Here is some code that "ties" an array to a file, so initially the array is "loaded" from the file, and changes to the array are reflected in the file too. This is done by opening it for appending, writing the changed data, and closing it - possibly slow, but very robust. If you don't give a filename, it is made up of the array name and the extension ".txt".

A special application for this is event logging: register a global persistent array once and assign to it, for instance with a telling key like clock seconds for each event:

persistentArray log mylogfile
...
set log([clock seconds]) "$this happened."
...
set log([clock seconds]) "$that happened."

When later examining the logfile, you can reconstruct the date and time of the events with clock format. The disadvantage against pure file logging is that all event messages remain in the array.

proc persistentArray {arrName {filename {}}} {
    upvar 1 $arrName arr
    array set arr {} ;# to make sure it exists, and is an array
    if {$filename==""} {set filename $arrName.txt}
    set filename [file join [pwd] $filename]
    if [file exists $filename] {
        set fp [open $filename]
        array set arr [read $fp]
        close $fp
    }
    uplevel 1 [list trace add variable $arrName {write unset} [list persist'save $filename]]
}
proc persist'save {filename arrName el op} {
    upvar 1 $arrName arr
    switch -- $op {
        write {set value $arr($el)}
        unset {set value {}}
    }
    set    fp [open $filename a]
    puts  $fp [list $el $value]
    close $fp
}

BerkeleyDB

Here's another approach to persistent/tied arrays that is similar to what perl does with dbmopen and whatnot. It uses BerkeleyDB.

# package require Db_tcl
load /usr/lib/libdb_tcl-3.2.so ;# already installed on some linux systems

proc tieArray {aname file db} {
    upvar $aname a
    array set a {}
    set fname [file nativename $file]
    set dbh [berkdb open -hash -create $fname $db]
    set tie_db::afiles($aname) $dbh
    trace add variable a read [list tie_db::read $aname]
    trace add variable a write [list tie_db::write $aname]
    trace add variable a unset [list tie_db::del $aname]
}

proc syncArray {aname} {
    $tie_db::afiles($aname) sync
}

proc syncAll {} {
    foreach a [array names tie_db::afiles] {
        syncArray $a
    }
}

proc untieArray {aname} {
    tie_db::afiles($aname) close
    trace remove variable read [list tie_db::read $aname]
    trace remove variable a write [list tie_db::write $aname]
    trace remove variable a unset [list tie_db::del $aname]
}


namespace eval ::tie_db {
    variable afiles

    proc read {rname aname ename op} {
        variable afiles
            upvar $aname a
            foreach {k v} [lindex [$afiles($rname) get $ename] 0] {null} {null} {}
            if {$k == ""} {
                error "no such element in db"
            }
            set a($ename) $v
    }

    proc write {rname aname ename op} {
        variable afiles
            upvar $aname a
            $afiles($rname) put $ename $a($ename)
            # $afiles($rname) sync
    }

    proc del {rname aname ename op} {
        variable afiles
            upvar $aname a
            $afiles($rname) del $ename 
            # $afiles($rname) sync
    }

    proc array {cmd aname args} {
        variable afiles
        if {[catch {set afiles($aname)}]} {
            uplevel 2 _array $cmd $aname $args
        } else {
            switch $cmd {
                names -
                get   -
                names/get {
                    set cur [$afiles($aname) cursor]
                    _array set tmp_array {}
                    while 1 {
                        set kvp [lindex [$cur get -next] 0]
                        if {$kvp == {}} {break}
                        puts -nonewline [lindex $kvp 0]..
                        set tmp_array([lindex $kvp 0]) [lindex $kvp 1]
                    }
                    $cur close
                    if {$cmd == "names"} {
                        _array names tmp_array
                    } elseif {$cmd == "get"} {
                        _array get tmp_array
                    }
                }
                default {error "$cmd not implemented for tied arrays"}
            }
        }
    }
}

rename array _array
proc array {args} {uplevel ::tie_db::array $args}

The write and del procedures can be changed to sync after each operation at a cost in performance. Some datasets seem to make this upset (I wrote it trying to make the Bayesan spam filtering code more elegant) but I suspect its some bugs in the version of db that I have.

Along with this, I noticed an anomaly with trace and upvar. You can use upvar to create an alias for a variable, even is the real variable is an element of an array.

% set c(x) 0
% upvar #0 c(x) x
% set x 1
% set c(x)
=> 1

However, that kind of aliased variable will not trigger traces set on the entire array, although it will trigger a trace on that element of the array. This seems like a bug; if a variable is traced then any way of accessing it should fire the same traces, right?

Tgdbm

Stefan Vogel 8-Mar-2004: You can also use the new version of Tgdbm (A Tcl-Wrapper for gdbm-(GNU-dbm)-API). Version 0.5 allows to "attach" an array to a gdbm-file (which stores the equivalent to the tcl-array, namely hash-key/value-pairs).

It is possible to do this:

gdbm_open -writer -sync -array airports test.gdbm
set airports(PAR) Paris ;# will store or update the key/value to test.gdbm

# add/update some more data
array set airports {
    ADD "Addis Abeba"
    FFM "Frankfurt"
}

# print value (gdbm-file and array is synchronized)
puts "FFM: [airports fetch FFM] / $airports(FFM)"
unset airports          ;# this will close test.gdbm
                        ;# this could have been done with unset airportArray

See http://www.vogel-nest.de (go to Tcl/Tgdbm) for more details on Tgdbm.

LMDB

dbohdan 2017-11-23: The following module tracks changes to an array using traces and persists them to an LMDB database.

Code

# Array persistence through LMDB.
# Copyright (c) 2017-2018 D. Bohdan
# License: MIT
package require Tcl 8.6
package require lmdb 0.3.5

namespace eval ::lmdbarr {
    variable version 0.2.1
}

::oo::class create ::lmdbarr::Persistent {
    variable _array
    variable _db
    variable _debug
    variable _env
    variable _path
    # If $_temporary is true, Persistent will delete the database under $_path
    # when destroyed.
    variable _temporary

    constructor {array {path {}} {temporary {}} {debug 0}} {
        if {[array size $array] > 0} {
            error "array must be empty when Persistent starts;\
                   $array has [array size $array] elements"
        }

        set _array $array
        set _path $path
        set _temporary $temporary
        set _debug $debug

        if {$_path eq {}} {
            close [file tempfile _path]
            file delete $_path
            file mkdir $_path
            if {$_temporary eq {}} {
                set _temporary 1
            }
        }
        set _env [lmdb env]
        $_env open -path $_path
        # Choose the map size: 1 TiB on 64-bit platforms and 2 GiB - 1 byte
        # otherwise.
        $_env set_mapsize [expr {
            $::tcl_platform(pointerSize) == 8 ? 1 << 40 : (1 << 31) - 1
        }]
        set _db [lmdb open -env $_env]
        my load
        my traces add
    }

    destructor {
        # Do nothing if the object wasn't initialized properly.
        if {![info exists _array]} {
            return
        }
        $_db close -env $_env
        $_env close
        if {$_temporary} {
            foreach file [glob -dir $_path *.mdb] {
                file delete $file
            }
            file delete $_path
        }
        my traces remove
    }

    method get-path {} {
        return $_path
    }

    method handle-trace {name1 name2 op} {
        if {$name1 ne $_array} return
        set txn [$_env txn]
        if {$_debug} {
            puts stderr [list handle-trace $name1 $name2 $op]
        }
        switch -exact -- $op {
            write {
                $_db put $name2 [set $_array\($name2\)] -txn $txn
            }
            unset {
                if {$name2 eq {}} {
                    # If the array has been unset, recreate the traces and empty
                    # the database.
                    my traces add
                    $_db drop 0 -txn $txn
                } else {
                    $_db del $name2 {} -txn $txn
                }
            }
        }
        $txn commit
        $txn close
    }

    method load {} {
        set txn [$_env txn]
        set cursor [$_db cursor -txn $txn]
        set initialValues {}
        try {
            while 1 {
                lappend initialValues {*}[$cursor get -next]
            }
        } on error {_ x} {
            # Ignore MDB_NOTFOUND errors.
            if {![string match {*ERROR: MDB_NOTFOUND*} \
                               [dict get $x -errorinfo]]} {
                return {*}$x
            }
        }
        array set $_array $initialValues
        $cursor close
        $txn commit
        $txn close
    }

    method set dict {
        my traces remove
        set txn [$_env txn]
        array set $_array $dict
        foreach {key value} $dict {
            $_db put $key $value -txn $txn
        }
        $txn commit
        $txn close
        my traces add
    }

    method traces op {
        uplevel [list trace $op variable $_array {write unset} \
                                [list [self namespace]::my handle-trace]]
    }

    method unset list {
        my traces remove
        set txn [$_env txn]
        foreach key $list {
            unset $_array\($key\)
            $_db del $key {} -txn $txn
        }
        $txn commit
        $txn close
        my traces add
    }

    unexport handle-trace load traces
}

proc ::lmdbarr::assert-equal {actual expected} {
    if {$actual ne $expected} {
        error "expected \"$expected\", but got \"$actual\""
    }
}

proc ::lmdbarr::test {} {
    # Tests.
    set sorted {varName {
        lsort -stride 2 -index 1 -integer [array get ::lmdbarr::$varName]
    }}
    set psr1 [Persistent new ::lmdbarr::arr {} 0]
    set path [$psr1 get-path]
    set ::lmdbarr::arr(false-start) -1
    unset ::lmdbarr::arr
    set ::lmdbarr::arr(foo) 2
    set ::lmdbarr::arr(bar) 3
    set ::lmdbarr::arr(baz) 5
    set ::lmdbarr::arr(qux) 7
    set ::lmdbarr::arr(quux) 11
    unset ::lmdbarr::arr(qux)
    array set ::lmdbarr::arr {baz 199}
    assert-equal [apply $sorted arr] {foo 2 bar 3 quux 11 baz 199}
    $psr1 destroy

    unset ::lmdbarr::arr
    set psr2 [Persistent new ::lmdbarr::arr $path 0]
    assert-equal [apply $sorted arr] {foo 2 bar 3 quux 11 baz 199}
    array unset ::lmdbarr::arr *
    array set ::lmdbarr::arr {x 0 y 1}
    assert-equal [apply $sorted arr] {x 0 y 1}
    $psr2 destroy
    unset ::lmdbarr::arr

    set psr3 [Persistent new ::lmdbarr::arr2 $path 1]
    assert-equal [apply $sorted arr2] {x 0 y 1}

    # Benchmark.
    set timex {{key script} {
        upvar time time
        dict lappend time $key [lindex [uplevel [list time $script]] 0]
    }}
    for {set run 0} {$run < 3} {incr run} {
        apply $timex set {
            for {set i 0} {$i < 1000} {incr i} {
                set ::lmdbarr::arr2(k$i) $i
            }
        }
        apply $timex get {
            for {set i 0} {$i < 1000} {incr i} {
                set ::lmdbarr::arr2(k$i)
            }
        }
        apply $timex unset {
            for {set i 0} {$i < 1000} {incr i} {
                unset ::lmdbarr::arr2(k$i)
            }
        }
        apply $timex mass-set {
            set d {}
            for {set i 0} {$i < 1000} {incr i} {
                dict set d k$i $i
            }
            $psr3 set $d
        }
        apply $timex mass-unset {
            set keys {}
            for {set i 0} {$i < 1000} {incr i} {
                lappend keys k$i
            }
            $psr3 unset $keys
        }
    }
    foreach {key value} $time {
        set sum 0
        foreach el $value {
            incr sum $el
        }
        puts [format "%-10s %6.1f ms" \
                     $key \
                     [expr {(1.0 * $sum) / [llength $value] / 1000}]]
    }
    $psr3 destroy
}

if {[info exists argv0] && ([file tail [info script]] eq [file tail $argv0])} {
    ::lmdbarr::test
}

Benchmark results

set        5035.5 ms
get           1.3 ms
unset      5078.7 ms
mass-set     11.6 ms
mass-unset   11.1 ms

Use example

# ex.tcl
source lmdbarr.tcl
set path /tmp/lmdbarr-example
file mkdir $path
set pst [::lmdbarr::Persistent new ::persist $path]
foreach {key value} $argv {
    set ::persist($key) $value
}
parray ::persist

$ tclsh ex.tcl 
$ tclsh ex.tcl k v
::persist(k) = v
$ tclsh ex.tcl 
::persist(k) = v
$ tclsh ex.tcl k2 v2
::persist(k)  = v
::persist(k2) = v2

Category Embedded Database

Category Example

Category File