Review Board has historically only supported Elasticsearch 1.x and 2.x.

This was due to limitations in Django Haystack, which didn't gain

anything more modern until the more recent 3.x releases. Unfortunately,

we can't depend on these releases, due to lack of Python 2.7 support and

a hard (install-time) requirement on Django 2.2.
In order to support these version, we've backported Haystack's

Elasticsearch backends into the Review Board tree. These are largely

compatible with Haystack 2.8, with the exception of some constants that

were missing (and easily redefined). They were also largely compatible

with Python 2.7, with the exception of some syntax errors (trailing

commas after a **kwargs in a function call).
Our Elasticsearch option will select the right backend based on the

installed version of the elasticsearch module. We display the

installed version on the configuration page, and provide instructions on

how to change the compatible version of Elasticsearch.
(There's also a small fix to the margins for fieldset descriptions, which

was very noticeable on this settings page.)
There's one key compatibility issue introduced by Elasticsearch 7

support, having to do with indexing. We were indexing lists of LocalSite

IDs as numeric, instead of strings. The field type would normally cast

to string, but that was overridden by us. This was fine until

Elasticsearch 7, which enforced stricter types, and wouldn't let us do a

query that factored in LocalSites the way we wanted. The solution is to

cast to a string.
This doesn't appear to regress anything when updating an index that

previously used longs (mixing the two types), and versions of

Elasticsearch prior to 7 don't appear to care. Still, we'll want to

recommend a full reindex on upgrade.

Unit tests pass on Python 2.7 and 3.x.
Manually tested this with the latest Elasticsearch 1.x, 2.x, 5.x, 7.x.

Each test involved:

Installing the appropriate build of elasticsearch (on Python 2.7

and 3.8).
Reloading the Search Settings page and verifying the version.
Updating the URL to point to an appropriate Elasticsearch server.
Rebuilding indexes with numeric Local Site IDs
Performing search queries
Rebuilding indexes with string Local Site IDs
Performing search queries

The numeric queries failed on Elasticsearch 7.x, as expected, but 7.x

users will be starting with a new index anyway.
Also tested incremental indexing on 1.x and 2.x, mixing numeric and

string Local Site IDs.