…his fixes #1590.

* Fix client over to xidmap

* Fix bug in shard index calculation

* Prefix keys in dgraph/client (since badger is used both for xids and checkpoints)

* Rename methods on xidmap

* Fix bug in lease file generation

* Allow users to specify performance options

* Add docs

* Don't require the uid provider implementation to accept size

* Fix off-by-one error in uid blocks

* Changes from self review

* Fix in checkpoint reading

* But more explicit about errors

* Review suggesions from Pawan (various)

- Query no longer contains mutations. This means:
  - No more query vars can be used in mutations.
  - No more upserts. They will be done via transactions.
  - Gql doesn't need to parse for mutations or schema mutations.
- Add a new endpoint called Mutate, just for mutations.
- Remove http endpoint /query. All read-write ops must be done via a grpc client.

2. Also move all the batching logic to dgraphloader. Users can just create the batch and set via SetObject.

- All responses would now be sent back as JSON bytes in response proto.
- Latency is part of response proto, not JSON output.

…xt PR

Add versioning for posting lists
vendor in latest badger.

* Get initial prototype working

* Check to make sure predicate is indexed correctly before using custom tokenizer

* Pass custom tokenizers in via flag

* Don't make plugin specify byte

* Add integer factorisation tokenizer (tokens are the prime factors of the int)

* Get tokenizer name from so filename

* Add example 'letter' tokenizer

* Add anagram example

* Use fixed size prefix for each custom tokenizer

* Ensure we don't get two tokenizers with the same hash

* Wrap error from custom tokenizer

* Add better cidr plugin

* Plugin now provides the full interface

* Don't require Tokenizer implementations to prepend identifier byte

* Check for valid tokenizer.Type() value

* Fix assert print

* Fix cidr

* Factor example (for int type)

* Convert to schema type before tokenizing in query

* Don't make implementors have IsSortable and IsLossy

* Fix rune example

* anagram plugin

* Don't attempt coverage report on plugins

* WIP for documentation (no examples yet)

* Code review suggestions from Manish

* Add system tests for plugins (needed to refactor systest to hook it in)

* Add examples to documentation

TODO:
1. Need to pass around readTs in worker/task.go
2. Change export, predicate_move

* Reduce layers and size of base dockerfile

pass back the keys changed in transaction

* Add fix for facet comparison bug for when types are mismatched

* Simplify float vs int comparison, and point out that it won't work for bigger values

- Add a lock over primary attribute.
- Add partial logic for commit or abort.
- Generate and propagate a transaction context which keeps track of keys and primary attribute.
- Client uses txn context for mutations, commit and abort.

Do commit/abort via proposals.
Restore primary key from disk.

    Test out the conflict fix logic during reads.
    Get primary and start ts both from list.
    Logic to fix conflict for a txn. Also pass read timestamp to all reads in worker/task.

…the keys.

* Check for "_nil_" for both DefaultID and StringID when generating output nodes

* Fix _nil_ in export

* Add test for export

…I now.

…ctly to update data in live Dgraph.

Don't set primary key in client
Remove touch index key
Write to disk before aborting/committing txn
Allow nquads in client, pass back the primary if already set.

…e is a simplified version of the previous loader code.

…there's an error. Various other conflict detection mechanisms to improve throughput.

Fix export, snapshot to use new iterator
Don't evict pl if it has pending mutation.
Special Handling for schema mutations.

* Disallow "uid" as argument for functions

* Add unit test