Mongos crash when we are moveChunk. Caused by the following log:
Tue Nov 12 20:01:12 [conn611979] Assertion: 10429:setShardVersion failed host: 10.38.171.25:7111

{ oldVersion: Timestamp 29317000|0, oldVersionEpoch: ObjectId('522ad499e1814e603d11be30'), ns: "appid250528.meta_infos0", version: Timestamp 2932 0000|0, versionEpoch: ObjectId('522ad499e1814e603d11be30'), globalVersion: Timestamp 29321000|0, globalVersionEpoch: Obje ctId('522ad499e1814e603d11be30'), reloadConfig: true, errmsg: "shard global version for collection is higher than trying to set to 'appid250528.meta_infos0'", ok: 0.0 }

When I use gdb debug the coredump file, something must be wrong. The details are:
1. chunk X move from shardA to shardB, chunkA's version update to 29363|0, and shardB's local version ChunkY update to 29363|1. And these updates will send to config server when moveChunk is finished.
2. When use gdb, the ChunkManager object of this ns is displayed as:
_version = 29363|1
_shardVersion[shardA] = 29363|1
_shardVersion[shardB] = 29362|0
3. So, when CheckShardVersion for shardB, since _version is the newest, the retry of 'conf->getChunkManager( ns , true )' will skip reload process.

The key reason is:
_version updated to the newest, but _shardVersion[shardB] is the old version.

I doubt :
When we update ChunkManage to version 29363|1, calculateConfigDiff read chunks info from config server , But first read the old version of ChunkX 29362|0, then the updates of step 1 happens, then read the new version of ChunkY 29363|1.

How to resolve:
when retry 3 times of 'conf->getChunkManager( ns , true )' in CheckShardVersion, use forceload for getChunkManager.

Hope your replies. Thank you!