| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
| |
The original cluster-wide tracking tables were transformed into local
tables, which are ram only. They can be moved into ets tables making
its use much simpler
|
| |\
| |
| | |
Adopt otp 23.3.4.16
|
| | | |
|
| |\ \
| | |
| | | |
Adopt otp 24.3.4.2
|
| | |/ |
|
| |\ \
| |/
|/| |
Adopt otp 25.0.3
|
| |/ |
|
| | |
|
| |\
| |
| | |
Include rabbit_index_route in rabbit_table:definitions/0
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This commit attempts to address the feedback in
https://github.com/rabbitmq/rabbitmq-server/pull/5254#discussion_r924497137.
Delete special code paths like function rabbit_table:ensure_feature_flag_tables/0
as they are confusing.
Instead, handle the new rabbit_index_route table as any other Mnesia
table if feature flag direct_exchange_routing_v2 is enabled.
Note that this commit comes with one important restriction:
A feature flag migration function MUST NOT call
rabbit_table:definitions/0 (directly or indirectly) as this will result
in a deadlock where the migration function is stuck on the global lock
with resource ID 'feature_flags_state_change' when checking for
rabbit_feature_flags:is_enabled(direct_exchange_routing_v2)
because the migration function runs in a different (RPC) process than
the function that holds the 'feature_flags_state_change' lock.
The latter lock is set before a feature gets enabled and released
after the feature flag got enabled.
|
| |\ \
| |/
|/| |
Unexpected fix for //deps/rabbitmq_shovel_management:http_SUITE
|
| | |
| |
| |
| |
| | |
I'm not sure why this change makes a difference, maybe an incorrect
compiler optimization for a record?
|
| |\ \
| |/
|/| |
CQ: Fix two accidental state overwrites
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The main issue was the one in fetch_by_predicate which resulted
in a double file:close on a single FD, the second failing and
triggering a badmatch. It only occurs for CQv2 in lazy mode
and with a high enough consume rate. It is possible that in
other conditions problems did occur but were not detected.
The other was detected while looking for the original issue,
but might not have created any bug.
|
| |\ \
| | |
| | | |
Move listeners from local ram mnesia table to ets tables
|
| | | | |
|
| |\ \ \
| |_|/
|/| | |
Split the OCI workflow into a nightly and on push phase for less flakiness
|
| | | |
| | |
| | |
| | |
| | | |
until the OCI Base workflow runs at least once, the custom base cannot
be used
|
| | | |
| | |
| | |
| | | |
So that flakiness of debug symbol installation do not block it
|
| | | |
| | |
| | |
| | |
| | | |
As they seem to be extremely flaky. They may be restored via alternate
approach.
|
| | | | |
|
| | | |
| | |
| | |
| | | |
in an attempt to reduce flakiness due to expired package mirrors
|
| |\ \ \
| |/ /
|/| | |
Log when acquiring or releasing a global lock
|
| | |/
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
to easily detect dead locks.
Prior to this commit, one caller of function
rabbit_ff_registry_factory:acquire_state_change_lock/0
was logging that it acquires the lock.
However, there are other callers which did not log.
Therefore, it was impossible to detect any dead locks by just
looking at the the logs.
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| |/
|
|
| |
Also automatically open PRs for adopting new OTP patch versions
|
| |\ |
|
| |/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Changes in this commit:
1.
Use feature_flags_v2 API for feature flag direct_exchange_routing_v2.
Both feature flags feature_flags_v2 and direct_exchange_routing_v2
will be introduced in 3.11.
Therefore, direct_exchange_routing_v2 can depend on feature_flags_v2.
2.
rabbit_binding:populate_index_route_table/0 should be run only during the
feature flag migration. Thereafter, upon node start-ups, binding recovery takes
care of populating the table.
3.
Merge direct_exchange_routing_v2_post_3_11_SUITE into
direct_exchange_routing_v2_SUITE. We don't need two separate test
suites with almost identical setup and teardown.
|
| |\
| |
| | |
Use win32sysinfo.exe to get disk and memory info
|
| | | |
|
| | |
| |
| |
| |
| | |
This is a patch by @ansd, and addresses
//deps/rabbit:direct_exchange_routing_v2_SUITE test run failures
|
| | | |
|
| | | |
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Follow-up to rabbitmq/rabbitmq-server#4140
In rabbitmq/rabbitmq-server#4328, I use Powershell to get disk info. This PR changes that
to use the "included with OTP" win32sysinfo.exe
Remove Powershell code
Correctly parse the return value of os_mon_sysinfo:get_disk_info/0
|
| |\ \
| | |
| | |
| | |
| | | |
rabbitmq/pjk25/less-noisy-memory-monitor-bazel-macos
Be less noisy when sysctl cannot be used for vm_memory_monitor
|
| |/ /
| |
| |
| |
| |
| |
| |
| | |
This avoids printing the full stacktrace when the error comes from the
sysctl invocation, the error message itself is sufficient
In practice, when testing with bazel with macos, sysctl is blocked by
the sandbox, so logging the stacktrace is rather noisy for tests
|
| | |
| |
| |
| |
| | |
This is a patch by @ansd, and addresses
//deps/rabbit:direct_exchange_routing_v2_SUITE test run failures
|
| |\ \
| |/
|/| |
|
| | | |
|
| |/ |
|
| |\
| |
| | |
Keep listeners on a local ram table
|
| | | |
|
| | | |
|
| |\ \
| | |
| | | |
Ensure rabbit_index_route table is created after joining a cluster
|
