Seed storage servers are recruited as the intial set of storage servers
when a database is first created. They function a little bit differently
than normal, and do not set an initial version like storages normally do
when they get recruited (typically equal to the recovery version).
Version correction is a feature where versions advance in sync with the
clock, and are equal across FDB clusters. To allow different FDB
clusters to have matching versions, they must share the same base
version. This defaults to the Unix epoch, and clusters with the version
epoch enabled will have a current version equal to the number of
microseconds since the Unix epoch.
When the version epoch is enabled on a cluster, it causes a one time
jump from the clusters current version to the version based on the
epoch. After a recovery, the recovery version sent to storages should
have advanced by a significant amount.
The recovery path contained a `BUGGIFY` to randomly advance the recovery
version in simulation, testing the version epoch being enabled.
However, it was also advancing the version during an initial recovery,
when the seed storage servers are recruited. If a set of storage
servers were recruited as seed servers, but another recovery occurred
before the bootstrap process was complete, the randomly selected version
increase could be smaller during the second recovery than during the
first. This could cause the initial set of seed servers to think they
should be at a version larger than what the cluuster was actually at.
The fix contained in this commit is to only cause a random version jump
when the recovery is occuring on an existing database, and not when it
is recruiting seed storages.
This commit fixes an issue found in simulation, reproducible with:
Commit: 93dc4bfeb9
Test: fast/DataLossRecovery.toml
Seed: 3101495991
Buggify: on
Compiler: clang
* Extend Tlog persistentStorage to persist encryption state
Description
diff-3: Address review comment.
diff-2: Extend ClusterController endpoints to allow query
cluster's encryptionAtRest status
Update Tlog recovery to ensure on-disk encryption
status matches with cluster's cstate persisted
encryptionAtRest
diff-1: Store encryptionAtRestMode state in Coordinators
Major changes proposed are:
1. Extend TLog persistentStorage to persist encryption state
2. Encryption state persisted is derived from corresponding
db-config and relevant SERVER_KNOBS. In near future, knobs
shall be removed.
3. On TLog startup, the persisted encryption state is compared
against cluster configuration, if mismatch, the TLog is killed
and not allowed to rejoin the cluster.
Testing
devRunCorrectness - 100K
The logic to determine the validity of a process joining a cluster now
belongs on the worker and the cluster controller. It is no longer
restricted to tlogs and storages, but instead applies to all processes
(even stateless ones).
Currently, there is a cyclic reference situation in
DatabaseContext -> WatchMetadata -> watchStorageServerResp ->
DatabaseContext
If there is a watch created in the DatabaseContext, even the
corresponding wait ACTOR is cancelled, the WatchMetadata will still hold
a reference to watchStorageServerResp ACTOR, which holds a reference to
DatabaseContext.
In this situation, any DatabaseContext who held a watch will not be
automatically destructed since its reference count will never reduce to
0 until the watch value is changed. Every time the cluster recoveries,
several watches are created, and when the cluster restarts, the
DatabaseContext which not being used, will not be able to destructed due
to these watches.
With this patch, each wait to the watch will be counted. Either the
watch is triggered or cancelled, the corresponding count will be
reduced. If a watch is not being waited, the watch will be cancelled,
effectively reduce the reference count of DatabaseContext. This will
hopefully fix the issue mentioned above.
The code is tested by 1) Manually change the number of logs of a local
cluster, see the cluster recovery and previous DatabaseContext being
destructed; 2) 100K joshua run, with 1 failure, the same test will fail
on the current git main branch.
The logic to determine the validity of a process joining a cluster now
belongs on the worker and the cluster controller. It is no longer
restricted to tlogs and storages, but instead applies to all processes
(even stateless ones).
Changes:
1. Change `isEncryptionOpSupported` to not check against `clientDBInfo.isEncryptionEnabled`, but instead against ENABLE_ENCRYPTION server knob. The problem with clientDBInfo is before its being broadcast to the workers, its content is uninitialized, during which some data (e.g. item 2) is not getting encrypted when they should.
2. Fix CommitProxy not encrypting metadata mutations which are recovered from txnStateStore
3. Fix KeyValueStoreMemory (thus TxnStateStore) partial transaction coming from recovery is not encrypted
4. new CODE_PROBE for the above fixes
5. Logging changes
* fix: the log router could see popped() versions between recoveredAt and the recovery txn version which cause remote tlogs to be stopped at a version in this range, which causes storage servers to read at versions in this range
* fix: a spurious correctness bug resulted from the consistency check running during a recovery
Configuration database data lives on the coordinators. When a change
coordinators command is issued, the data must be sent to the new
coordinators to keep the database consistent.
* 'main' of github.com:sfc-gh-nwijetunga/foundationdb: (32 commits)
Store rocksdb::DBOptions and rocksdb::ColumnFamilyOptions to (#7766)
Update CONTRIBUTING.md
Update tests/rare/SpecificUnitTests.toml
fix ASAN OOM problem
Update CONTRIBUTING.md
Write tracing and ALP special key errors as JSON
Fix: the static tenant map in the Java tester was being accessed concurrently from multiple threads. Make it a concurrent map. (#7805)
Run clang-format
Print SIGNAL output to stdout
Print to stderr only upon errors
Testing upgrades to a future version of FDB (#7780)
Flush gcov coverage upon SIGTERM
Report the unit tests being run in test harness
Fix a bug in a storage wiggler unit test where some servers were added with too recent a timestamp
Fix undefined behavior in versioned btree test due to integer overflow
When a transaction operation gets an unknown tenant error, it needs to reset the tenant ID so it can be updated in the next tenant lookup request.
Don't buggify max tenants per cluster globally; instead buggify it in specific tests
Remove non-existing unittest
Add unit tests to the correctness package
Add comment to INetwork
...
* Enable configuring the next future protocol version as the current protocol version in FDB client, fdbserver, and fdbcli
* Auto format python files used in upgrade tests
* Add a test for upgrading to a future FDB version
* Emphasize that the options for using future protocol version are intended for test purposes only
* Make the global variable for current protocol version visible only locally
* Refactirng to avoid using currentProtocolVersion() in static intialization
* Update go bindings
There was a rare but possible issue where a change coordinators request
could be in flight, but reading `\xff/coordinators` would return the old
set of coordinators, even though a previous attempt to write it returned
`commit_unknown_result`. This happened if there was an unrelated
recovery between setting `\xff/coordinators` and the updated
coordinators reaching each machine. The `commit_unknown_result` caused
by this recovery meant the change coordinators retry loop would read
`\xff/coordinators`, but since the request was ongoing it would read the
old set of coordinators. The fix is to have the cluster controller
ignore any change coordinators request from an old generation, meaning
when the retry loop reads the old coordiantors from `\xff/coordinators`,
it is guaranteed that the in progress change coordinators message will
be rejected when arriving at the cluster controller.
* proof of concept
* use code-probe instead of test
* code probe working on gcc
* code probe implemented
* renamed TestProbe to CodeProbe
* fixed refactoring typo
* support filtered output
* print probes at end of simulation
* fix missed probes print
* fix deduplication
* Fix refactoring issues
* revert bad refactor
* make sure file paths are relative
* fix more wrong refactor changes
Adding encryption support for TxnStateStore. It is done by supporting encryption. for KeyValueStoreMemory. The encryption is currently done on operation level when the operations are being write to the underlying log file. See inline comment for the encrypted data format.
This PR depends on #7252. It is part of the effort to support TLog encryption #6942.
Zhe Wu