[BUG] locking issue in cache.go `fatal error: concurrent map writes`

[mcarbonne]

Describe the bug

When running the latest version of OCIS, a SW relying on go-micro, I have some crashes (owncloud/ocis#10785)

The backtrace look like this (I extracted only the relevant part):

fatal error: concurrent map writes
fatal error: concurrent map writes

goroutine 13004 [running]:
go-micro.dev/v4/registry/cache.(*cache).isValid(0xc003746c60, {0xc0099c38c0, 0x1, 0x16?}, {0xc009750fa0?, 0x0?, 0x5f85c8271d40?})
        go-micro.dev/v4@v4.11.0/registry/cache/cache.go:103 +0x1aa
go-micro.dev/v4/registry/cache.(*cache).get(0xc003746c60, {0x5f85c3fe03b1, 0x16})
        go-micro.dev/v4@v4.11.0/registry/cache/cache.go:145 +0x136
go-micro.dev/v4/registry/cache.(*cache).GetService(0x5f85c3fe03b1?, {0x5f85c3fe03b1?, 0x5f85c5856f90?}, {0x1?, 0x0?, 0xc00a766b80?})
        go-micro.dev/v4@v4.11.0/registry/cache/cache.go:462 +0x18
github.com/cs3org/reva/v2/pkg/rgrpc/todo/pool.(*Selector[...]).Next(0x5f85c6241710, {0x0, 0x0, 0x16})
        github.com/cs3org/reva/v2@v2.27.0/pkg/rgrpc/todo/pool/selector.go:117 +0x2b4

I did some research and in cache.go I've the following code:

func (c *cache) get(service string) ([]*registry.Service, error) {
	// read lock
	c.RLock()

	// check the cache first
	services := c.cache[service]
	// get cache ttl
	ttl := c.ttls[service]
	// make a copy
	cp := util.Copy(services)

	// got services, nodes && within ttl so return cache
	if c.isValid(cp, ttl) {
		c.RUnlock()
		// return services
		return cp, nil
	}

From what I understood, this function take a "read only lock" but isValid need a RW lock, because of this line:

delete(c.nttls, s.Name)

The bug might have been introduced with this PR #2736

[mcarbonne]

Isn't this fix OK ?

[asim]

Have you tested since that change?