... | ... | @@ -26,7 +26,7 @@ We do encourage people to test this and report their experience. We're particula |
|
|
|
|
|
**Multi-threaded HA work**. Kea 1.8 introduced multi-threaded support for DHCP packet processing, which offers substantial performance gains. Unfortunately, if used with HA, the current implementation makes the lease updates sequential, effectively eliminating the benefits of multi-threading. This limitation requires substantial rework of how HA connections are established. In this milestone we made substantial progress towards solving this bottleneck. The MT-HA design has been proposed and reviewed #1315. Also, the first major element of the solution (a multi-threaded HTTP listener) has been reviewed and merged #1748. The HA-MT solution is not functional yet, but definitely completed the experiments and planning phase and moved onto the implementation phase. This work will continue in the next releases.
|
|
|
|
|
|
**Test farm migration**. While this is not something users are normally concerned with, it's an aspect that took a substantial amount of our team's time. Our Jenkins test farm was running on ISC hosted hardware that was aging and was not very powerful. We often struggled with test execution times and were limited in the number of systems we could test on. Our QA team set up a new Jenkins instance using AWS (Amazon Web Services). Kea test environment is exceedingly complicated (with MySQL, PostgreSQL, Cassandra backends, Radius, NETCONF, DNS servers, multiple Kea instances running in a variety of stand-alone, HA, load-balancing modes, and various API and traffic generators). We managed to migrate the great majority of our environment. While the bulk of the migration is done, not everything is running smoothly on the new platform yet. As such, our testing of this release is more lightweight than usual. It doesn't mean the release is broken or has lower quality than usual. It simply means that we were not able to confirm its stability. Use with caution! #1774, #1761, #1773, #1770, #1751, #1757, #1752
|
|
|
**Test farm migration**. While this is not something users are normally concerned with, it's an aspect that took a substantial amount of our team's time. Our Jenkins test farm was running on ISC hosted hardware that was aging and was not very powerful. We often struggled with test execution times and were limited in the number of systems we could test on. Our QA team set up a new Jenkins instance using AWS (Amazon Web Services). Kea test environment is exceedingly complicated (with MySQL, PostgreSQL, Cassandra backends, RADIUS, NETCONF, DNS servers, multiple Kea instances running in a variety of stand-alone, HA, load-balancing modes, and various API and traffic generators). We managed to migrate the great majority of our environment. While the bulk of the migration is done, not everything is running smoothly on the new platform yet. As such, our testing of this release is more lightweight than usual. It doesn't mean the release is broken or has lower quality than usual. It simply means that we were not able to confirm its stability. Use with caution! #1774, #1761, #1773, #1770, #1751, #1757, #1752
|
|
|
|
|
|
**Design for incremental lease updates**. The Stork project would like to be able to provide more insight into leases being assigned. However, the current API is not well prepared for it. It can enumerate all leases, but using existing calls periodically would incur a substantial burden on a Kea server. As such, a design for cached mode and incremental lease updates has been written. #1230
|
|
|
|
... | ... | |