SearchVarnish Enterprise 6.0.18r3 is a feature and bug fix release. It contains important bug fixes and stability improvements, and all users are encouraged to upgrade.
Connections to backends using TLS are now reused without having to re-issue a full TLS handshake.
The Memory Governor can now enter the overloaded state when the configured
memory_target is exceeded and it is not able to remove objects to reduce
memory usage. This can be observed in VCL to react and the strict_governor
feature flag can be enabled to prevent further object creations.
Improved alignment of memory allocations internally for fetches used during a
pass. This in particular is relevant for users of the transit_buffer.
For MSE4 the MSE4_NewObject processed bytes log record has now a more precise
value only counting the bytes that were processed instead of allocated for
processing.
headerplus: New case_match parameter can be used when matching
attribute names to choose case sensitivity of the matching.
mse4: New .is_gov_overloaded() function allows checking if the
Memory Governor is overloaded or not in VCL.
mse: New .is_gov_overloaded() function allows checking if the
Memory Governor is overloaded or not in VCL.
std: The std.fileread() function now stores per-VCL information,
making it possible to control when the content is updated.
cluster.vcl: A deadlocks could occur when subrequests went through
the cluster as they bypassed the self-routing loop prevention mechanism.
vmod_slicer: The slicer now always relies on the object internal
information instead of relying on VCL-editable headers.
vmod_image: Improved validation and clearer error message during
transformations.