Varnish Enterprise 6.0.18r8 is a bug fix release.
It is based on 6.0.18r6 and does not include the VCL compiler update from 6.0.18r7, which was removed. That update will return in a later release. Third-party VMODs built for 6.0.18r6 work with this release.
If you installed 6.0.18r7, VCL that was modified to use features new in 6.0.18r7 must be reverted, and third-party VMODs must be rebuilt for 6.0.18r6, before upgrading to this release.
Fixed several crashes in the cache: a grace object banned or evicted
during a lookup could still be delivered (VS issue #3540), a
connection arriving during socket.close or while the cache process
was stopping could crash it (VS issue #2327), and an object’s storage
could be freed twice during eviction (VS issue #3539).
nodes no
longer panics the cache process by delivering a configuration update
to a cold VCL. A cold VCL catches up when it warms again.
(VS issue #3581)
MSE4: fixed a race in the slot journal that could panic the cache
process on aarch64 (VS issue #3498). Fetches no longer wait forever
when a book runs out of free slots: the new slot_alloc_timeout book
parameter, 5 seconds by default, bounds the wait (VS issue #3499).
Piped requests, CONNECT tunnels and HTTP/2 sessions over TLS could stall when a TLS record carried more data than the internal read buffer.
Decreasing http_max_hdr at runtime could panic the cache process
when a request started a backend fetch or a sub-request.
(VS issue #3257)
See the changelog for the complete list of changes.
