In progress: Create store API to enable non-json based stores #9

Open
Ghost wants to merge 3 commits from (deleted):js-stores into httpd
1 changed files with 331 additions and 62 deletions
Showing only changes of commit e43c169257 - Show all commits

View File

@ -1,10 +1,99 @@
'use strict';
var crypto = require('crypto');
function jsonDeepClone(target) {
return JSON.parse(
JSON.stringify(target)
);
}
function mergeObjects() {
// arguments should be an array of objects. We
// reverse it because the last argument to set
// a value wins.
var args = [].slice.call(arguments).reverse();
var len = args.length;
if (len === 1) {
return args[0];
}
// gather the set of keys from all arguments
var keyLists = args.map(function (arg) {
return Object.keys(arg);
});
var keys = Object.keys(keyLists.reduce(function (all, list) {
list.forEach(function (k) {
all[k] = true;
});
return all;
}, {}));
// for each key
return keys.reduce(function (target, k) {
// find the first argument (because of the reverse() above) with the key set
var values = [];
var isObject = false;
for (var i = 0; i < len; i++) {
var v = args[i];
var vType = typeof v;
if (vType === 'object') {
if (!v) {
// typeof null is object. null is the only falsey object. null represents
// a delete or the end of our argument list;
break;
}
// we need to collect values until we get a non-object, so we can merge them
values.push(v);
isObject = true;
} else if (!isObject) {
if (vType === 'undefined') {
// if the arg actually has the key set this is effectively a "delete"
if (keyList[i].indexOf(k) != -1) {
break;
}
// otherwise we need to check the next argument's value, so we don't break the loop
} else {
values.push(v);
break;
}
} else {
// a previous value was an object, this one isn't
// That means we are done collecting values.
break;
}
}
if (values.length > 0) {
target[k] = mergeObjects.apply(null, values);
}
return target;
}, {});
}
function prepareZone(zone, options) {
var opts = options || {};
var timestamp = opts.timestamp || Date.now();
if (!zone.name) {
zone.name = zone.id;
zone.id = null;
}
if (!zone.id) {
zone.id = crypto.randomBytes(16).toString('hex');
}
if (!zone.createdAt) { zone.createdAt = timestamp; }
if (!zone.updatedAt || opts.isUpdate) { zone.updatedAt = timestamp; }
// create a names set for the zone, keyed by record name mapped to
// an object for the various records with that name, by type (A, MX, TXT, etc.)
zone.records = zone.records || {};
return zone;
}
/*
init() should return an object with: {
save: function -> undefined - changes to in memory representation should be persisted
@ -14,14 +103,12 @@ init() should return an object with: {
},
zones: {
list: function -> list zones,
find: function -> read zone by ???,
create:
update:
delete:
},
records: {
list: function -> list records,
find: function -> read record by ???,
create:
update:
delete:
@ -35,51 +122,71 @@ module.exports = function init (opts) {
// opts = { filepath };
var db = require(opts.filepath);
var stat = require('fs').statSync(opts.filepath);
var crypto = require('crypto');
var mtime = require('fs').statSync(opts.filepath).mtime.valueOf();
//
// Manual Migration
// Migration from other formats
//
// Convert the primary nameservers from strings to objects with names and IDs.
db.primaryNameservers.forEach(function (ns, i, arr) {
if ('string' === typeof ns) {
ns = { name: ns };
arr[i] = ns;
}
if (!ns.id) {
ns.id = crypto.randomBytes(16).toString('hex');
}
// Convert the primary nameservers from an array of strings to objects with names and IDs.
// also switch to the 'peers' name, since we are really interested in the other FQDNs that
// use the same data store and are kept in sync.
var peerList = (!db.peers || Array.isArray(db.peers))? db.peers : Object.keys(db.peers).map(function (p) {
return db.peers[p];
});
db.peers = [].concat(db.primaryNameservers, peerList).filter(function (p) {
// filer out empty strings, undefined, etc.
return !!p;
}).map(function (ns) {
var peer = ('string' === typeof ns)? ns : { name: ns };
if (!ns.id) {
peer.id = crypto.randomBytes(16).toString('hex');
}
return peer;
}).reduce(function (peers, p) {
peers[p.name] = p;
return peers;
}, {});
delete db.primaryNameservers;
// Convert domains to zones and ensure that they have proper IDs and timestamps
db.zones = db.zones || [];
if (db.domains) {
db.zones = db.zones.concat(db.domains);
}
db.zones.forEach(function (zone) {
if (!zone.name) {
zone.name = zone.id;
zone.id = null;
}
if (!zone.id) {
zone.id = crypto.randomBytes(16).toString('hex');
}
if (!zone.createdAt) { zone.createdAt = stat.mtime.valueOf(); }
if (!zone.updatedAt) { zone.updatedAt = stat.mtime.valueOf(); }
// Organize zones as a set of zone names
var zoneList = (!db.zones || Array.isArray(db.zones))? db.zones : Object.keys(db.zones).map(function (z) {
return db.zones[z];
});
// Records belong to zones, but they (currently) refer to them by a zone property.
// NOTE/TODO: This may pose problems where the whole list of records is not easily
db.zones = [].concat(db.domains, zoneList).filter(function (z) {
// filer out empty strings, undefined, etc.
return !!z;
}).map(function (zone) {
return prepareZone(zone, { timestamp: mtime });
}).reduce(function (zones, z) {
zones[z.name] = z;
return zones;
}, {});
delete db.domains;
// NOTE: Records belong to zones, but they previously referred to them only by a
// zone property. This may pose problems where the whole list of records is not easily
// filtered / kept in memory / indexed and/or retrieved by zone. Traditionally,
// records are stored "within a zone" in a zone file. We may wish to have the
// DB API behave more traditionally, even though some stores (like a SQL database
// records are stored "within a zone" in a zone file. We want to have the store API
// behave more traditionally, even though some stores (like a SQL database
// table) might actually store the zone as a property of a record as we currently do.
db.records.forEach(function (record) {
// (This fits with the somewhat unexpected and confusing logic of wildcard records.)
(db.records || []).forEach(function (record) {
// make sure the record has an ID
if (!record.id) {
record.id = crypto.randomBytes(16).toString('hex');
}
// put it in it's zone - synthesize one if needed
db.zones[record.zone] = db.zones[record.zone] || prepareZone({ name: record.zone });
var zone = db.zones[record.zone];
zone.records[record.name] = zone.records[record.name] || [];
var recordsForName = zone.records[record.name];
recordsForName.push(record);
});
delete db.records;
// Write the migrated data
require('fs').writeFileSync(opts.filepath, JSON.stringify(db, null, 2));
@ -95,7 +202,6 @@ module.exports = function init (opts) {
}
save._saving = true;
// TODO: replace with something not destructive to original non-json data
require('fs').writeFile(opts.filepath, JSON.stringify(db, null, 2), function (err) {
console.log('done writing');
var pending = save._pending.splice(0);
@ -112,54 +218,217 @@ module.exports = function init (opts) {
};
save._pending = [];
function matchPredicate(predicate) {
return function (toCheck) {
// which items match the predicate?
if (!toCheck) {
return false;
}
// check all the keys in the predicate - only supporting exact match
// of at least one listed option for all keys right now
if (Object.keys(predicate || {}).some(function (k) {
return [].concat(predicate[k]).indexOf(toCheck[k]) === -1;
})) {
return false;
}
// we have a match
return true;
};
}
function matchZone(predicate) {
var zonenames = !!predicate.name ? [].concat(predicate.name) : Object.keys(db.zones);
var check = matchPredicate(predicate);
// TODO: swap the filter() for a functional style "loop" recursive function
// that lets us return early if we have a limit, etc.
var found = zonenames.filter(function (zonename) {
/*
if (predicate.id && predicate.id !== z.id) { return false; }
if (predicate.name && predicate.name !== z.name) { return false; }
*/
return check(db.zones[zonename]);
}).map(function (zonename) {
return db.zones[zonename];
});
return found;
}
var dbApi = {
save: function () {
// hide _pending and _saving from callers
var args = [].slice.call(arguments);
return save.apply(null, args);
},
// primaryNameservers really isn't editable - it's literally the list of FQDN's
// peers really isn't editable - it's literally the list of FQDN's
// that this database is replicated to in a multi-master fashion.
//
// However, lib/store/index.js does plenty to update these records in support
// of the SOA records that are built from them (as does this file in the "migration"
// section). I'm toying with the idea of not storing them seperately or creating the
// SOA records somewhat immediately.
primaryNameservers: {
list: function listNameservers() {
return jsonDeepClone(db.primaryNameservers);
}
peers: function listPeers(cb) {
// Most data stores are going to have an asynchronous storage API. If we need
// synchronous access to the data it is going to have to be cached. If it is
// cached, there is still the issue the cache getting out of sync (a legitimate
// issue anyway). If we explicitly make all of these operations async then we
// have greater flexibility for store implmentations to address these issues.
return setImmediate(cb, null, jsonDeepClone(db.peers));
},
zones: {
list: function listZones() {
return jsonDeepClone(db.zones);
},
find: function getZone(predicate, cb) {
var found;
db.zones.some(function (z) {
if (z.id && predicate.id === z.id) { found = z; return true; }
if (z.name && predicate.name === z.name) { found = z; return true; }
/*
I'm fairly certan that zone names must be unique and therefore are legitimately
IDs within the zones namespace. This is similarly true of record names within a zone.
I'm not certain that having a distinct ID adds value and it may add confusion / complexity.
*/
// NOTE: `opts` exists so we can add options - like properties to read - easily in the future
// without modifying the function signature
list: function listZones(predicate, opts, cb) {
// TODO: consider whether we should just return the zone names
var found = jsonDeepClone(matchZone(predicate)).map(function (z) {
// This is fairly inefficient!! Consider alternative storage
// that does not require deleting the records like this.
delete z.records;
return z;
});
if (!found) {
cb(null, null);
return;
}
cb(null, jsonDeepClone(found));
return;
return setImmediate(cb, null, found);
},
create: function() {},
update: function() {},
delete: function() {}
// // NOTE: I'm not sure we need a distinct 'find()' operation in the API
// // unless we are going to limit the output of the
// // 'list()' operation in some incompatible way.
// // NOTE: `opts` exists so we can add options - like properties to read - easily in the future
// // without modifying the function signature
// find: function getZone(predicate, opts, cb) {
// if (!predicate.name || predicate.id) {
// return setImmediate(cb, new Error('Finding a zone requires a `name` or `id`'));
// }
// // TODO: implement a limit / short circuit and possibly offset
// // to allow for paging of zone data.
// var found = matchZone(predicate);
// if (!found[0]) {
// // TODO: make error message more specific?
// return setImmediate(cb, new Error('Zone not found'));
// }
// var z = jsonDeepClone(found[0]);
// delete z.records;
// return setImmediate(cb, null, z);
// },
create: function createZone(zone, cb) {
// We'll need a lock mechanism of some sort that works
// for simultaneous requests and multiple processes.
matchZone({ name: zone.name }, function (err, matched) {
if (err) {
return setImmediate(cb, err);
}
var found = matched[0];
if (found) {
return setImmediate(cb, new Error('Zone ' + zone.name + ' already exists'));
}
db.zones[zone.name] = prepareZone(zone);
return setImmediate(function () {
cb(null, jsonDeepClone(db.zones[zone.name]));
// release lock
});
});
},
update: function updateZone(zone, cb) {
// We'll need a lock mechanism of some sort that works
// for simultaneous requests and multiple processes.
matchZone({ name: zone.name }, function (err, matched) {
if (err) {
return setImmediate(cb, err);
}
var found = matched[0];
if (!found) {
return setImmediate(cb, new Error('Zone not found'));
}
// make sure we are not writing records through this interface
delete zone.records;
var combined = mergeObjects(found, zone);
db.zones[zone.name] = prepareZone(combined, { isUpdate: true });
return setImmediate(function () {
cb(null, jsonDeepClone(db.zones[zone.name]));
// release lock
});
});
},
delete: function(zone, cb) {
// We'll need a lock mechanism of some sort that works
// for simultaneous requests and multiple processes.
matchZone({ name: zone.name }, function (err, matched) {
if (err) {
return setImmediate(cb, err);
}
var found = matched[0];
if (!found) {
return setImmediate(cb, new Error('Zone not found'));
}
delete db.zones[zone.name];
return setImmediate(function () {
cb();
// release lock
});
});
}
},
records: {
list: function listRecords() {
return jsonDeepClone(db.records);
list: function listRecords(rPredicate, cb) {
var recordNames = [].concat(rPredicate.name);
var check = matchPredicate(rPredicate);
var found = matchZone({ name: rPredicate.zone }).reduce(function (records, zone) {
// get the records from the zone that match the record predicate
var zFound = recordNames.filter(function (name) {
return !!zone.records[name];
}).map(function (name) {
return zone.records[name].filter(check);
});
return records.concat(zFound);
}, []);
return setImmediate(cb, null, jsonDeepClone(found));
},
find: function getRecord(predicate, cb) {
// find: function getRecord(rPredicate, cb) {
// var recordNames = [].concat(rPredicate.name);
// var check = matchPredicate(rPredicate);
// // TODO: swap the `filter()` and `map()` for a functional style "loop"
// // recursive function that lets us return early if we have a limit, etc.
// var found = matchZone({ name: rPredicate.zone }).reduce(function (records, zone) {
// // get the records from the zone that match the record predicate
// var zFound = recordNames.filter(function (name) {
// return !!zone.records[name];
// }).map(function (name) {
// return zone.records[name].filter(check);
// });
// return records.concat(zFound);
// }, []);
// return setImmediate(cb, null, jsonDeepClone(found[0]));
// },
create: function(record, cb) {
var zone = matchZone({ name: record.zone })[0];
if (!zone) {
return setImmediate(cb, new Error('Unble to find zone ' + record.zone + ' to create record'));
}
var records = zone.records[record.name] = zone.records[record.name] || [];
var check = matchPredicate(record);
if (records.filter(check)[0]) {
return setImmediate(cb, new Error('Exact record already exists in zone ' + record.zone ));
}
return setImmediate(cb, null, jsonDeepClone(found));
},
create: function() {},
update: function() {},
delete: function() {}
update: function(record, cb) {},
delete: function(record, cb) {}
}
};