Concurrent map access

[meirf]

It looks like tree_cache's map is susceptible to concurrent read/write.

fatal error: concurrent map read and map write
goroutine 26 [running]:
runtime.throw(0xaf4cd7, 0x21)
    /usr/local/go/src/runtime/panic.go:596 +0x95 fp=0xc42007da38 sp=0xc42007da18
runtime.mapaccess2_faststr(0xa164a0, 0xc4201f4b70, 0xc4201425f0, 0xd, 0x0, 0x0)
    /usr/local/go/src/runtime/hashmap_fast.go:326 +0x50a fp=0xc42007da98 sp=0xc42007da38
fsgo/net/discovery.(*TreeCache).readInstanceList(0xc4201abf80, 0xc4201425f0, 0xd, 0xe2f6e0, 0x0, 0x0)
   fsgo/net/discovery/tree_cache.go:55 +0xb1 fp=0xc42007dc70 sp=0xc42007da98
fsgo/net/discovery.(*TreeCache).readAndWatch(0xc4201abf80, 0xc4201425f0, 0xd, 0xae2425, 0x8)
fsgo/net/discovery/tree_cache.go:48 +0x31e fp=0xc42007dd40 sp=0xc42007dc70
fsgo/net/discovery.(*TreeCache).readServices(0xc4201abf80, 0xc4201f4ba0)
fsgo/net/discovery/tree_cache.go:116 +0x2e8 fp=0xc42007df80 sp=0xc42007dd40
fsgo/net/discovery.(*TreeCache).processServiceChanges.func1(0xc4201abf80, 0xc4201f4ba0)
fsgo/net/discovery/tree_cache.go:97 +0x54 fp=0xc42007dfd0 sp=0xc42007df80
runtime.goexit()
    /usr/local/go/src/runtime/asm_amd64.s:2197 +0x1 fp=0xc42007dfd8 sp=0xc42007dfd0
fsgo/net/discovery.(*TreeCache).processServiceChanges
fsgo/net/discovery/tree_cache.go:100 +0x7d

goroutine 1 [runnable, locked to thread]:
fsgo/net/discovery.(*TreeCache).readInstanceList(0xc4201abf80, 0xc42001b810, 0xd, 0xe2f6e0, 0x0, 0x0)
	fsgo/net/discovery/tree_cache.go:84 +0x8ca
fsgo/net/discovery.(*TreeCache).readAndWatch(0xc4201abf80, 0xc42001b810, 0xd, 0xae2425, 0x8)
	fsgo/net/discovery/tree_cache.go:48 +0x31e
fsgo/net/discovery.(*TreeCache).readServices(0xc4201abf80, 0xc4201f4ba0)
	fsgo/net/discovery/tree_cache.go:116 +0x2e8
fsgo/net/discovery.(*TreeCache).processServiceChanges(0xc4201abf80)
	fsgo/net/discovery/tree_cache.go:101 +0x95
fsgo/net/discovery.(*TreeCache).Start(0xc4201abf80)
	fsgo/net/discovery/tree_cache.go:25 +0x2b

[meirf]

I think what happened is if the watcher detects an service change quickly it can update the list during the initial load.

[meirf]

Do you guys have an idea of how to fix this? This is causing us some pretty nasty errors.
Thanks!