region_sharding: working resharding example

[deepthi]

This modifies the region_sharding example to demonstrate region_json based resharding.

• We start with an unsharded keyspace
• Load some sample data
• Create a sharded vschema using region_json vindex
• Create and externalize a lookup vindex
• Bring up target shards (4 equal shards)
• Reshard
• SwitchReads and SwitchWrites
• Bring down shard 0

Note: DO NOT MERGE, I'm updating the user guide on the website vitessio/website#488, and would like to merge both at the same time.

Signed-off-by: deepthi deepthi@planetscale.com

[jmoldow]

In the original example on https://vitess.io/docs/user-guides/region-sharding/ , we also had to define the primary vindex on the lookup table:

    "customer_lookup": {
      "column_vindexes": [
	{
            "column": "keyspace_id",
            "name": "identity"
        }
      ]
    },

My suggestion on Slack was that we needed to change that to a primary vindex on id instead.

Is that missing from this example? Or are you making the lookup table unsharded? At the moment I don't see any column_vindexes defined for customer_lookup.

Is there another PR with the changes to the tutorial document https://vitess.io/docs/user-guides/region-sharding/ ?

I'm curious to see a discussion in the tutorial on the sharding strategy for the lookup table.

• If we shard by id (which makes the most sense from a functionality point-of-view; otherwise the lookup table is useless), or if we make the table unsharded, then we might still have to go cross-region for the lookups, which will add latency.
• Unless we cross-replicate every shard into every region, but then there are questions about replication lag consistency when using the lookup table.

[deepthi]

I'm using the new workflows that weren't available when we first built the example. CreateLookupVindex also creates a primary vindex on the lookup table.
This is what the vschema looks like after the lookup vindex is created.

$ vtctlclient GetVSchema main
{
  "sharded": true,
  "vindexes": {
    "customer_region_lookup": {
      "type": "consistent_lookup_unique",
      "params": {
        "from": "id",
        "table": "main.customer_lookup",
        "to": "keyspace_id"
      },
      "owner": "customer"
    },
    "hash": {
      "type": "hash"
    },
    "region_vdx": {
      "type": "region_json",
      "params": {
        "region_bytes": "1",
        "region_map": "/path/to/file/countries.json"
      }
    }
  },
  "tables": {
    "customer": {
      "columnVindexes": [
        {
          "name": "region_vdx",
          "columns": [
            "id",
            "country"
          ]
        },
        {
          "column": "id",
          "name": "customer_region_lookup"
        }
      ]
    },
    "customer_lookup": {
      "columnVindexes": [
        {
          "column": "id",
          "name": "hash"
        }
      ]
    }
  }
}

I'm still working on rewriting the user guide, PR should be ready shortly.

[deepthi]

Do you think it is helpful for CreateLookupVindex to return the created vschema? Otherwise it is silent and not obvious.

[jmoldow]

I haven't run vitess yet or gotten familiar with the vtctlclient commands, so I'm not in a good position to say. But I'll agree that it is not obvious, and finding ways to make it more obvious would probably be very helpful.