Added Agency API #104
Added Agency API #104
Conversation
| // agencyConnectionFailureBackoff returns a backoff delay for cases where all | ||
| // agents responded with a non-fatal error. | ||
| func agencyConnectionFailureBackoff(lastDelay time.Duration) time.Duration { | ||
| return increaseDelay(lastDelay, 1.5, time.Millisecond, time.Second*2) |
There was a problem hiding this comment.
I fear that a maximum delay of 2 seconds will not be enough in cases in which the machine is under a heavy load. I suggest to increase the maximum considerably to maybe 15 or 30 seconds.
There was a problem hiding this comment.
I don't understand why. The result of this function is used as a delay period between failing requests. It is not the timeout for an agency request.
There was a problem hiding this comment.
OK, then I misunderstood this, sorry. Please ignore.
| cancel() | ||
| close(results) | ||
| close(errors) | ||
| if result, ok := <-results; ok { |
There was a problem hiding this comment.
I do not understand what happens if a 307 is returned. Is this reported as an error? Or does each request follow the redirect? Both would be bad. I think if we send to all agents we need to ignore 307, but I do not understand how this works here.
There was a problem hiding this comment.
The underling HTTP connection is configured to fail in redirect. That means a 302 error is returned. The code here will consider any non-permanent failure (everything in the 300 range) as "this agent did not handle, so we'll wait for another".
Only in case all agents did not respond with either a result or a permanent failure, we'll retry the operation.
So the notion of automatic retry does not exist in any other case.
| // WriteKey writes the given value with the given key with a given TTL (unless TTL is zero). | ||
| // If you pass a condition (only 1 allowed), this condition has to be true, | ||
| // otherwise the write will fail with a ConditionFailed error. | ||
| func (c *agency) WriteKey(ctx context.Context, key []string, value interface{}, ttl time.Duration, condition ...WriteCondition) error { |
There was a problem hiding this comment.
In this function an automatic retry is OK, since it is idempotent.
| } | ||
|
|
||
| // WriteKeyIfEmpty writes the given value with the given key only if the key was empty before. | ||
| func (c *agency) WriteKeyIfEmpty(ctx context.Context, key []string, value interface{}, ttl time.Duration) error { |
There was a problem hiding this comment.
Here, a retry can be disastrous, since the operation is not idempotent. This is actually a very difficult problem. There is the possibility to send a client ID along with a transaction, such that one is later able to enquire if an operation has actually happened already. However, this makes the client code quite difficult to get right (see our AgencyComm, we had considerable trouble to implement this correctly, LOL, and I am not sure we got it right now). Furthermore, even that is not entirely water-tight because of the rather fundamental two generals problem. We never know whether the first request is sent of and lingers in some TCP buffer, and is sent after a longish delay. During the time when it is in the buffers, it is not possible to either stop it or to find out if it will be executed eventually. I do not know what the right approach is here.
|
|
||
| // WriteKeyIfEqualTo writes the given new value with the given key only if the existing value for that key equals | ||
| // to the given old value. | ||
| func (c *agency) WriteKeyIfEqualTo(ctx context.Context, key []string, newValue, oldValue interface{}, ttl time.Duration) error { |
|
|
||
| // RemoveKeyIfEqualTo removes the given key only if the existing value for that key equals | ||
| // to the given old value. | ||
| func (c *agency) RemoveKeyIfEqualTo(ctx context.Context, key []string, oldValue interface{}) error { |
This PR adds an Agency API.
Note that this API is not intended for normal database use.