Probably related: Tencent/rapidjson#388

A suggested work-around is switching RapidJSON to CrtAllocator on ARM, which probably wouldn't be the worst thing (it might cost a little bit of performance).

Note if someone wants to try this out:
To change the allocator, open the file src/rapidjson/document.h. Then change the following two lines:

• line 420 from
  template <typename Encoding, typename Allocator = MemoryPoolAllocator<> > class GenericValue {
  to
  template <typename Encoding, typename Allocator = RAllocator > class GenericValue {
• and line 1626 from
  template <typename Encoding, typename Allocator = MemoryPoolAllocator<>, typename StackAllocator = RAllocator> class GenericDocument : public GenericValue<Encoding, Allocator> {
  to
  template <typename Encoding, typename Allocator = RAllocator, typename StackAllocator = RAllocator> class GenericDocument : public GenericValue<Encoding, Allocator> {

After making the changes, do I need to do anything else beside recompile? It takes a long to recompile on Cubietruck so I want to make sure I do not miss any step.

@paulscliu That should be all.

I made the modification and recompile. RethinkDB 2.1.4 now run fine on the Cubietruck. I will let it run for few days to see if there is any problem but I don't anticipate any

This solved the problem for me as well. I'm testing on a couple of https://www.scaleway.com/ C1 servers.

Thanks for the feedback @pauliusuza and @chrj .
Scheduling this for 2.2. We'll try to incorporate this work-around into our codebase, so that everything runs out of the box again.

I think you meant @paulscliu

Oh sorry, yes I meant @paulscliu

The server does run on the beaglebone black.

It now ignores the --server-name switch from the command line as well as the same setting line from the config file.

WARNING: ignoring --server-name because this server already has a name.

I'm not sure what the reason could be, it seems like all 2.1.x versions do it.

@10man: --server-name only sets a name if the server does not have one (i.e.: is starting for the first time). Once there is a data folder created and populated that then directs what the server name is. You can change it through the admin interface (WebGUI: "Operations" drop-down(s) on the "Servers" page) or with a ReQL command like:

r.db('rethinkdb').table('server_config').get(UUID_HERRE).update({'name':'NEW NAME'}).run(conn)

There was some in-person conversation in the office about whether a server should refuse to start when given a --server-name that does not match what it has, but a number of good arguments about cases where that would cause outsized problems convinced me that that was not a good idea (e.g.: settings files used for cluster setup).

A fix for this is in code review 3260 by @encryptio and will probably be included in RethinkDB 2.1.5.

The work-around has been merged into v2.1.x 9b8dbc2 and will ship with RethinkDB 2.1.5.
Also forward-ported to next (including changing term_storage to not use the MemoryPoolAllocator on ARM) with 851b2f7.