You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
PR #371 introduced that long polling subscriptions would always fetch the
next n (defaulted to 3) long polling indices to reduce the latency from
roundtrip to single trip.
For the happy path this works well, but in case of errors, the logic has
to be improved to correctly handle responses arriving out of order.
With the current logic this will re-fetch already received updates and
lead to avalanches of requests, as these re-fetched requests will then
also again arive out of order and refetch old data.
Before increasing this value again, it should be ensured that the
responses are either re-sorted on arrival or older updates correctly
dropped without scheduling new requests.
This change restores the behaviour from before the changes but keeps the
logic in place so it can be reenabled after fixing the error-handling.
Signed-off-by: Alexander Krimm <[email protected]>
staticconstexpr std::size_tkParallelLongPollingRequests = 1; // increasing this value could reduce latency but needs some more robust error handling for unexpected updates
constexpr std::size_tkParallelLongPollingRequests = 1; // increasing this value could reduce latency but needs some more robust error handling for unexpected updates
0 commit comments