Using compression to store up to 10x the normal amount of data in a cell

I recently came up with a useful tool for my Airtable toolkit, and I thought I would share it in case others may find useful.

On a recent project, I needed to transform a large amount of data into a JSON object for use by external applications. Unfortunately, I quickly ran up against Airtable’s 100,000 character limit on text fields. It occurred to me that if I could find a way to compress the JSON data, I could fit more data into a cell and wouldn’t have to worry about the character limit.

There are a handful of javascript libraries out there for text compression, but what I settled on using is the lz-string.js library. It is based on LZW compression, and was designed for use on mobile browsers, meaning that it is fast, and optimized enough to use in Airtable Automations without going over the CPU time caps. Due to its repetitive nature, JSON data compresses very well—results will vary depending on the structure of your data, but with my current data I’m able to fit the equivalent of 900,000 characters in a cell, rather than 100,000.

Consider this a proof of concept—if you need to access this data from outside of Airtable you’ll need to find a way to decompress it (FWIW, this library has been ported to a number of different languages). There are some other javascript compression libraries out there that may be more interoperable, but for my use case this was the easiest to set up. I’m using this data for generating printed documents on demand, so I do the decompression from within an app.

I posted a base to the Airtable Universe if anyone wants to give it a whirl, unfortunately the scripts don’t appear to be included though :tired_face:. I implemented the scripts as automations so that I could benchmark them, but it’s just as easy to implement them in an app. Here are the scripts, I trigger them off of the “Run Compression” and “Run Decompression” checkboxes. Just clear out all the cells to the right of the “Data” column to see the results of running the compression scripts yourself. Oh, and make sure to set up an input variable on the automation with the key recordID and the Record ID as value.

Compression

/**
 * LZ-String
 * For more information, the home page:
 * http://pieroxy.net/blog/pages/lz-string/testing.html
 * 
 * LZ-based compression algorithm, version 1.4.4
 */
var LZString=function(){function o(o,r){if(!t[o]){t[o]={};for(var n=0;n<o.length;n++)t[o][o.charAt(n)]=n}return t[o][r]}var r=String.fromCharCode,n="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/=",e="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+-$",t={},i={compressToBase64:function(o){if(null==o)return"";var r=i._compress(o,6,function(o){return n.charAt(o)});switch(r.length%4){default:case 0:return r;case 1:return r+"===";case 2:return r+"==";case 3:return r+"="}},decompressFromBase64:function(r){return null==r?"":""==r?null:i._decompress(r.length,32,function(e){return o(n,r.charAt(e))})},compressToUTF16:function(o){return null==o?"":i._compress(o,15,function(o){return r(o+32)})+" "},decompressFromUTF16:function(o){return null==o?"":""==o?null:i._decompress(o.length,16384,function(r){return o.charCodeAt(r)-32})},compressToUint8Array:function(o){for(var r=i.compress(o),n=new Uint8Array(2*r.length),e=0,t=r.length;t>e;e++){var s=r.charCodeAt(e);n[2*e]=s>>>8,n[2*e+1]=s%256}return n},decompressFromUint8Array:function(o){if(null===o||void 0===o)return i.decompress(o);for(var n=new Array(o.length/2),e=0,t=n.length;t>e;e++)n[e]=256*o[2*e]+o[2*e+1];var s=[];return n.forEach(function(o){s.push(r(o))}),i.decompress(s.join(""))},compressToEncodedURIComponent:function(o){return null==o?"":i._compress(o,6,function(o){return e.charAt(o)})},decompressFromEncodedURIComponent:function(r){return null==r?"":""==r?null:(r=r.replace(/ /g,"+"),i._decompress(r.length,32,function(n){return o(e,r.charAt(n))}))},compress:function(o){return i._compress(o,16,function(o){return r(o)})},_compress:function(o,r,n){if(null==o)return"";var e,t,i,s={},p={},u="",c="",a="",l=2,f=3,h=2,d=[],m=0,v=0;for(i=0;i<o.length;i+=1)if(u=o.charAt(i),Object.prototype.hasOwnProperty.call(s,u)||(s[u]=f++,p[u]=!0),c=a+u,Object.prototype.hasOwnProperty.call(s,c))a=c;else{if(Object.prototype.hasOwnProperty.call(p,a)){if(a.charCodeAt(0)<256){for(e=0;h>e;e++)m<<=1,v==r-1?(v=0,d.push(n(m)),m=0):v++;for(t=a.charCodeAt(0),e=0;8>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1}else{for(t=1,e=0;h>e;e++)m=m<<1|t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t=0;for(t=a.charCodeAt(0),e=0;16>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1}l--,0==l&&(l=Math.pow(2,h),h++),delete p[a]}else for(t=s[a],e=0;h>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1;l--,0==l&&(l=Math.pow(2,h),h++),s[c]=f++,a=String(u)}if(""!==a){if(Object.prototype.hasOwnProperty.call(p,a)){if(a.charCodeAt(0)<256){for(e=0;h>e;e++)m<<=1,v==r-1?(v=0,d.push(n(m)),m=0):v++;for(t=a.charCodeAt(0),e=0;8>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1}else{for(t=1,e=0;h>e;e++)m=m<<1|t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t=0;for(t=a.charCodeAt(0),e=0;16>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1}l--,0==l&&(l=Math.pow(2,h),h++),delete p[a]}else for(t=s[a],e=0;h>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1;l--,0==l&&(l=Math.pow(2,h),h++)}for(t=2,e=0;h>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1;for(;;){if(m<<=1,v==r-1){d.push(n(m));break}v++}return d.join("")},decompress:function(o){return null==o?"":""==o?null:i._decompress(o.length,32768,function(r){return o.charCodeAt(r)})},_decompress:function(o,n,e){var t,i,s,p,u,c,a,l,f=[],h=4,d=4,m=3,v="",w=[],A={val:e(0),position:n,index:1};for(i=0;3>i;i+=1)f[i]=i;for(p=0,c=Math.pow(2,2),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;switch(t=p){case 0:for(p=0,c=Math.pow(2,8),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;l=r(p);break;case 1:for(p=0,c=Math.pow(2,16),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;l=r(p);break;case 2:return""}for(f[3]=l,s=l,w.push(l);;){if(A.index>o)return"";for(p=0,c=Math.pow(2,m),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;switch(l=p){case 0:for(p=0,c=Math.pow(2,8),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;f[d++]=r(p),l=d-1,h--;break;case 1:for(p=0,c=Math.pow(2,16),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;f[d++]=r(p),l=d-1,h--;break;case 2:return w.join("")}if(0==h&&(h=Math.pow(2,m),m++),f[l])v=f[l];else{if(l!==d)return null;v=s+s.charAt(0)}w.push(v),f[d++]=s+v.charAt(0),h--,s=v,0==h&&(h=Math.pow(2,m),m++)}}};return i}();"function"==typeof define&&define.amd?define(function(){return LZString}):"undefined"!=typeof module&&null!=module&&(module.exports=LZString);

// ********************

let config = input.config();
let recordID = config.recordID;
let table = base.getTable('LZ-String Compression');
let query = await table.selectRecordsAsync();
let record = query.getRecord(recordID);
let data = record.getCellValue('Data');

let compressed = LZString.compress(data);

await table.updateRecordAsync(recordID, {
    "Compressed Data": compressed,
    "Run Compression": false
});

Decompression

/**
 * LZ-String
 * For more information, the home page:
 * http://pieroxy.net/blog/pages/lz-string/testing.html
 * 
 * LZ-based compression algorithm, version 1.4.4
 */
var LZString=function(){function o(o,r){if(!t[o]){t[o]={};for(var n=0;n<o.length;n++)t[o][o.charAt(n)]=n}return t[o][r]}var r=String.fromCharCode,n="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/=",e="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+-$",t={},i={compressToBase64:function(o){if(null==o)return"";var r=i._compress(o,6,function(o){return n.charAt(o)});switch(r.length%4){default:case 0:return r;case 1:return r+"===";case 2:return r+"==";case 3:return r+"="}},decompressFromBase64:function(r){return null==r?"":""==r?null:i._decompress(r.length,32,function(e){return o(n,r.charAt(e))})},compressToUTF16:function(o){return null==o?"":i._compress(o,15,function(o){return r(o+32)})+" "},decompressFromUTF16:function(o){return null==o?"":""==o?null:i._decompress(o.length,16384,function(r){return o.charCodeAt(r)-32})},compressToUint8Array:function(o){for(var r=i.compress(o),n=new Uint8Array(2*r.length),e=0,t=r.length;t>e;e++){var s=r.charCodeAt(e);n[2*e]=s>>>8,n[2*e+1]=s%256}return n},decompressFromUint8Array:function(o){if(null===o||void 0===o)return i.decompress(o);for(var n=new Array(o.length/2),e=0,t=n.length;t>e;e++)n[e]=256*o[2*e]+o[2*e+1];var s=[];return n.forEach(function(o){s.push(r(o))}),i.decompress(s.join(""))},compressToEncodedURIComponent:function(o){return null==o?"":i._compress(o,6,function(o){return e.charAt(o)})},decompressFromEncodedURIComponent:function(r){return null==r?"":""==r?null:(r=r.replace(/ /g,"+"),i._decompress(r.length,32,function(n){return o(e,r.charAt(n))}))},compress:function(o){return i._compress(o,16,function(o){return r(o)})},_compress:function(o,r,n){if(null==o)return"";var e,t,i,s={},p={},u="",c="",a="",l=2,f=3,h=2,d=[],m=0,v=0;for(i=0;i<o.length;i+=1)if(u=o.charAt(i),Object.prototype.hasOwnProperty.call(s,u)||(s[u]=f++,p[u]=!0),c=a+u,Object.prototype.hasOwnProperty.call(s,c))a=c;else{if(Object.prototype.hasOwnProperty.call(p,a)){if(a.charCodeAt(0)<256){for(e=0;h>e;e++)m<<=1,v==r-1?(v=0,d.push(n(m)),m=0):v++;for(t=a.charCodeAt(0),e=0;8>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1}else{for(t=1,e=0;h>e;e++)m=m<<1|t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t=0;for(t=a.charCodeAt(0),e=0;16>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1}l--,0==l&&(l=Math.pow(2,h),h++),delete p[a]}else for(t=s[a],e=0;h>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1;l--,0==l&&(l=Math.pow(2,h),h++),s[c]=f++,a=String(u)}if(""!==a){if(Object.prototype.hasOwnProperty.call(p,a)){if(a.charCodeAt(0)<256){for(e=0;h>e;e++)m<<=1,v==r-1?(v=0,d.push(n(m)),m=0):v++;for(t=a.charCodeAt(0),e=0;8>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1}else{for(t=1,e=0;h>e;e++)m=m<<1|t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t=0;for(t=a.charCodeAt(0),e=0;16>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1}l--,0==l&&(l=Math.pow(2,h),h++),delete p[a]}else for(t=s[a],e=0;h>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1;l--,0==l&&(l=Math.pow(2,h),h++)}for(t=2,e=0;h>e;e++)m=m<<1|1&t,v==r-1?(v=0,d.push(n(m)),m=0):v++,t>>=1;for(;;){if(m<<=1,v==r-1){d.push(n(m));break}v++}return d.join("")},decompress:function(o){return null==o?"":""==o?null:i._decompress(o.length,32768,function(r){return o.charCodeAt(r)})},_decompress:function(o,n,e){var t,i,s,p,u,c,a,l,f=[],h=4,d=4,m=3,v="",w=[],A={val:e(0),position:n,index:1};for(i=0;3>i;i+=1)f[i]=i;for(p=0,c=Math.pow(2,2),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;switch(t=p){case 0:for(p=0,c=Math.pow(2,8),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;l=r(p);break;case 1:for(p=0,c=Math.pow(2,16),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;l=r(p);break;case 2:return""}for(f[3]=l,s=l,w.push(l);;){if(A.index>o)return"";for(p=0,c=Math.pow(2,m),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;switch(l=p){case 0:for(p=0,c=Math.pow(2,8),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;f[d++]=r(p),l=d-1,h--;break;case 1:for(p=0,c=Math.pow(2,16),a=1;a!=c;)u=A.val&A.position,A.position>>=1,0==A.position&&(A.position=n,A.val=e(A.index++)),p|=(u>0?1:0)*a,a<<=1;f[d++]=r(p),l=d-1,h--;break;case 2:return w.join("")}if(0==h&&(h=Math.pow(2,m),m++),f[l])v=f[l];else{if(l!==d)return null;v=s+s.charAt(0)}w.push(v),f[d++]=s+v.charAt(0),h--,s=v,0==h&&(h=Math.pow(2,m),m++)}}};return i}();"function"==typeof define&&define.amd?define(function(){return LZString}):"undefined"!=typeof module&&null!=module&&(module.exports=LZString);

// ********************

let config = input.config();
let recordID = config.recordID;
let table = base.getTable('LZ-String Compression');
let query = await table.selectRecordsAsync();
let record = query.getRecord(recordID);
let data = record.getCellValue('Compressed Data');

let decompressed = LZString.decompress(data);

await table.updateRecordAsync(recordID, {
    "Decompressed Data": decompressed,
    "Run Decompression": false
});
3 Likes

Very cool! Is there much of a performance hit when decompressing?

1 Like

The majority of the performance hit happens when compressing, decompression is very fast. The compression overhead is significant enough that you’ll need to be cautious about it, but decompression shouldn’t be an issue. I made a benchmark just now with a JSON string of 436k characters and these were the results. Compression used ~170ms of CPU time (after subtracting the CPU time for everything else in the script), and decompression used ~40ms of CPU time.

Note that the data here is randomized dummy data generated with https://next.json-generator.com/, so it’s an idealized situation as regards compression ratio. Under normal circumstances, you’re not going to get quite so good compression, unless your data has a lot of repetition.

As an aside, if you need to go beyond the 100,000 character limit you could just split your data across multiple columns and stitch it back together when you need to use it, but the compression method is cooler :sunglasses:

3 Likes