* Harden CommitteeIndex, SubnetId, SyncSubcommitteeIndex
Harden the use of `CommitteeIndex` et al to prevent future issues by
using a distinct type, then validating before use in several cases -
datatypes in spec are kept simple though so that invalid data still can
be read.
* fix invalid epoch used in REST
`/eth/v1/beacon/states/{state_id}/committees` committee length (could
return invalid data)
* normalize some variable names
* normalize committee index loops
* fix `RestAttesterDuty` to use `uint64` for `validator_committee_index`
* validate `CommitteeIndex` on ingress in REST API
* update rest rules with stricter parsing
* better REST serializers
* save lots of memory by not using `zip` ...at least a few bytes!
* REST cleanups
* reject out-of-range committee requests
* print all hex values as lower-case
* allow requesting state information by head state root
* turn `DomainType` into array (follow spec)
* `uint_to_bytesXX` -> `uint_to_bytes` (follow spec)
* fix wrong dependent root in `/eth/v1/validator/duties/proposer/`
* update documentation - `--subscribe-all-subnets` is no longer needed
when using the REST interface with validator clients
* more fixes
* common helpers for dependent block
* remove test rules obsoleted by more strict epoch tests
* fix trailing commas
* Update docs/the_nimbus_book/src/rest-api.md
* Update docs/the_nimbus_book/src/rest-api.md
Co-authored-by: sacha <sacha@status.im>
Used during initial development of the spec, `nbench` has fallen behind
and by and large been superceded by `block_sim`, `state_sim` and
`ncli_db bench`.
* ncli_db: add putState, putBlock
These tools allow modifying an existing nimbus database for the purpose
of recovery or reorg, moving the head, tail and genesis to arbitrary
points.
* remove potentially expensive `putState` in `BeaconStateDB`
* introduce `latest_block_root` which computes the root of the latest
applied block from the `latest_block_header` field (instead of passing
it in separately)
* avoid some unnecessary BeaconState copies during init
* discover https://github.com/nim-lang/Nim/issues/19094
* prefer `HashedBeaconState` in a few places to avoid recomputing state
root
* fetch latest block root from state when creating blocks
* harden `get_beacon_proposer_index` against invalid slots and document
* move random spec function tests to `test_spec.nim`
* avoid unnecessary state root computation before block proposal
* Support starting from altair
* hide `finalized-checkpoint-` - they are incomplete and usage may cause
crashes
* remove genesis detection code (broken, obsolete)
* enable starting ChainDAG from altair checkpoints - this is a
prerequisite for checkpoint sync (TODO: backfill)
* tighten checkpoint state conditions
* show error when starting from checkpoint with existing database (not
supported)
* print rest-compatible JSON in ncli/state_sim
* altair/merge support in ncli
* more altair/merge support in ncli_db
* pre-load header to speed up loading
* fix forked block decoding
* Address review comments in #3057
* reorder imports in rest_utils
maybe this will help with the mysterious serialization issues
Co-authored-by: Jacek Sieka <jacek@status.im>
According to the spec, this call should return the positions of
the specified validators within the sync committee. The existing
code was instead returning the indices of the sync sub-committees
where the validator is a member.
* fix stack overflow crash in REST/debug/getStateV2
* introduce `ForkyXxx` for generic type matching of `Xxx` across
branches (SomeHashedBeaconState -> ForkyHashedBeaconState et al) -
`Some` is already used for other types of type classes
* consolidate function naming in BeaconChainDB, use some generics
* import `forks.nim` from other spec modules and move `Forked*` helpers
around to resolve circular imports
* remove `ForkedBeaconState`, use `ForkedHashedBeaconState` throughout
(less data shuffling between the types)
* fix several cases of states being stored on stack in tests, causing
random failures on some platforms
* remove reading json support from ncli - this should be ported to the
rest json reading instead (doesn't currently work because stack sizes)
Similar to the existing `RewardInfo`, this PR adds the infrastructure
needed to export epoch processing information from altair+. Because
accounting is done somewhat differently, the PR uses a fork-specific
object to extrct the information in order to make the cost on the spec
side low.
* RewardInfo -> EpochInfo, ForkedEpochInfo
* use array for computing new sync committee
* avoid repeated total active balance computations in block processing
* simplify proposer index check
* simplify epoch transition tests
* pre-compute base increment and reuse in epoch processing, and a few
other small optimizations
This PR introduces the type and does the heavy lifting in terms of
refactoring - the tools that use the accounting will need separate PR:s
(as well as refinements to the exportred information)
* "official" -> "scenarios", like the submodule
* fewer test binaries - various compile hacks have been improved over
time, test suite should follow
* remove obsolete bls tests - there are better test vectors in
nim-blscurve
* remove obsolete mentions of `ssz_testing`
* remove obsolete comments about proc vs globals, unittest2 already uses
proc's
* add merge SSZ consensus object tests
* add merge attestations, attester slashing, proposer slashing, and voluntary exit test fixtures
* throw catchable exception rather than assert on invalid SingleMemberUnion selector
* hash_tree_root can assert, because it's trusted data by that point
* re-assert on writing, since also trusted data
* beta.4 update
* implement upgrade_to_merge()
* construct all unslashed, participating balances in one validator scan
* remove altair benchmarking setup for block_sim
* revert a benchmarking change from proc to func
* remove more benchmarking func/proc tweaks
* re-add asSeq, which is necessary because unsafeAddr
* ... except for the block_sim benchmark part
* approximately double altair epoch processing speed
* don't redundantly clear state balances cache
* only invalidate altair state balances cache once in process_rewards_and_penalties()
* reorganize ssz dependencies
This PR continues the work in
https://github.com/status-im/nimbus-eth2/pull/2646,
https://github.com/status-im/nimbus-eth2/pull/2779 as well as past
issues with serialization and type, to disentangle SSZ from eth2 and at
the same time simplify imports and exports with a structured approach.
The principal idea here is that when a library wants to introduce SSZ
support, they do so via 3 files:
* `ssz_codecs` which imports and reexports `codecs` - this covers the
basic byte conversions and ensures no overloads get lost
* `xxx_merkleization` imports and exports `merkleization` to specialize
and get access to `hash_tree_root` and friends
* `xxx_ssz_serialization` imports and exports `ssz_serialization` to
specialize ssz for a specific library
Those that need to interact with SSZ always import the `xxx_` versions
of the modules and never `ssz` itself so as to keep imports simple and
safe.
This is similar to how the REST / JSON-RPC serializers are structured in
that someone wanting to serialize spec types to REST-JSON will import
`eth2_rest_serialization` and nothing else.
* split up ssz into a core library that is independendent of eth2 types
* rename `bytes_reader` to `codec` to highlight that it contains coding
and decoding of bytes and native ssz types
* remove tricky List init overload that causes compile issues
* get rid of top-level ssz import
* reenable merkleization tests
* move some "standard" json serializers to spec
* remove `ValidatorIndex` serialization for now
* remove test_ssz_merkleization
* add tests for over/underlong byte sequences
* fix broken seq[byte] test - seq[byte] is not an SSZ type
There are a few things this PR doesn't solve:
* like #2646 this PR is weak on how to handle root and other
dontSerialize fields that "sometimes" should be computed - the same
problem appears in REST / JSON-RPC etc
* Fix a build problem on macOS
* Another way to fix the macOS builds
Co-authored-by: Zahary Karadjov <zahary@gmail.com>
The spec imports are a mess to work with, so this branch cleans them up
a bit to ensure that we avoid generic sandwitches and that importing
stuff generally becomes easier.
* reexport crypto/digest/presets because these are part of the public
symbol set of the rest of the spec types
* don't export `merge` types from `base` - this causes circular deps
* fix circular deps in `ssz/spec_types` - this is the first step in
disentangling ssz from spec
* be explicit about phase0 vs altair - longer term, `altair` will become
the "natural" type set, then merge and so on, so no point in giving
`phase0` special preferential treatment
Simpler module name for stuff that covers forks
* check that runtime config matches database state
* also include some assorted altair cleanups
* use "standard" genesis fork in local testnet to work around missing
runtime config support
* Implement split preset/config support
This is the initial bulk refactor to introduce runtime config values in
a number of places, somewhat replacing the existing mechanism of loading
network metadata.
It still needs more work, this is the initial refactor that introduces
runtime configuration in some of the places that need it.
The PR changes the way presets and constants work, to match the spec. In
particular, a "preset" now refers to the compile-time configuration
while a "cfg" or "RuntimeConfig" is the dynamic part.
A single binary can support either mainnet or minimal, but not both.
Support for other presets has been removed completely (can be readded,
in case there's need).
There's a number of outstanding tasks:
* `SECONDS_PER_SLOT` still needs fixing
* loading custom runtime configs needs redoing
* checking constants against YAML file
* yeerongpilly support
`build/nimbus_beacon_node --network=yeerongpilly --discv5:no --log-level=DEBUG`
* load fork epoch from config
* fix fork digest sent in status
* nicer error string for request failures
* fix tools
* one more
* fixup
* fixup
* fixup
* use "standard" network definition folder in local testnet
Files are loaded from their standard locations, including genesis etc,
to conform to the format used in the `eth2-networks` repo.
* fix launch scripts, allow unknown config values
* fix base config of rest test
* cleanups
* bundle mainnet config using common loader
* fix spec links and names
* only include supported preset in binary
* drop yeerongpilly, add altair-devnet-0, support boot_enr.yaml
* remove false OnBlockAdded dependency on phase.HashedBeaconState
* introduce altair data types into block_clearance; update some alpha.6 spec refs to alpha.7; add get_active_validator_indices_len ForkedHashedBeaconState wrapper
* switch many modules from using datatypes (with phase0 states/blocks) to datatypes/base (fork-independent); update spec refs from alpha.6 to alpha.7 and remove rm'd G2_POINT_AT_INFINITY
* switch more modules from using datatypes (with phase0 states/blocks) to datatypes/base (fork-independent); update spec refs from alpha.6 to alpha.7
* remove unnecessary phase0-only wrapper of get_attesting_indices(); allow signatures_batch to process either fork; remove O(n^2) nested loop in process_inactivity_updates(); add altair support to getAttestationsforTestBlock()
* add Altair versions of asSigVerified(), asTrusted(), and makeBeaconBlock()
* fix spec URL to be Altair for Altair makeBeaconBlock()
* update to Altair as of v1.1.0-alpha.7
* introduce Altair types into attestation pool
* avoid allocating/copying pubkeys excessively in get_next_sync_committee()
* use ForkedHashedBeaconState in StateData
* fix FAR_FUTURE_EPOCH -> slot overflow; almost always use assign()
* avoid stack allocation in maybeUpgradeStateToAltair()
* create and use dispatch functions for check_attester_slashing(), check_proposer_slashing(), and check_voluntary_exit()
* use getStateRoot() instead of various state.data.hbsPhase0.root
* remove withStateVars.hashedState(), which doesn't work as a design anymore
* introduce spec/datatypes/altair into beacon_chain_db
* fix inefficient codegen for getStateField(largeStateField)
* state_transition_slots() doesn't either need/use blocks or runtime presets
* combine process_slots(HBS)/state_transition_slots(HBS) which differ only in last-slot htr optimization
* getStateField(StateData, ...) was replaced by getStateField(ForkedHashedBeaconState, ...)
* fix rollback
* switch some state_transition(), process_slots, makeTestBlocks(), etc to use ForkedHashedBeaconState
* remove state_transition(phase0.HashedBeaconState)
* remove process_slots(phase0.HashedBeaconState)
* remove state_transition_block(phase0.HashedBeaconState)
* remove unused callWithBS(); separate case expression from if statement
* switch back from nested-ref-object construction to (ref Foo)(Bar())
Instead of keeping a validator key list per EpochRef, this PR introduces
a single shared validator key list in ChainDAG, and cleans up some other
ChainDAG and key-related issues.
The PR does not introduce the validator key list in the state transition
- this is because we batch-check all signatures before entering the spec
code, thus the spec code never hits the cache.
A future refactor should _probably_ remove the threadvar altogether.
There's a few other small fixes in here that make the flow easier to
read:
* fix `var ChainDAGRef` -> `ChainDAGRef`
* fix `var QuarantineRef` -> `QuarantineRef`
* consistent `dag` variable name
* avoid using threadvar pubkey cache in most cases
* better error messages in batch signature checking
* proposed structure for hf1
* refactor datatypes.nim into datatypes/{base, phase0, hf1}.nim
* hf1 is Altair
* some syncing with alpha 2
* adjust epoch processing to disambiguate access to RewardFlags
* relocate StateData to stay consistent with meaning phase 0 StateData
* passes v1.1.0 alpha 5 SSZ consensus object tests
* Altair block header test fixtures work
* fix slash_validator() so that Altair attester slashings, proposer slashings, and voluntary exit textures work
* deposit operation Altair test fixtures work
* slot sanity and all but a couple epoch transition tests switched to Altair
* attestation Altair test fixtures work
* Altair block sanity test fixtures work
* add working altair sync committee tests
* improve workarounds for sum-types-across-modules Nim bug; incorporate SignedBeaconBlock root reconstuction to SSZ byte reader
* use StateData in place of BeaconState outside state transition code
* propagate more StateData usage
* remove withStateVars().state
* wrap get_beacon_committee(BeaconState, ...) as gbc(StateData, ...)
* switch makeAttestation() to use StateData
* use StateData wrapper/dispatcher for get_committee_count_per_slot()
* convert AttestationCache.init(), weak subjectivity functions, and updateValidatorMetrics()
* add get_shuffled_active_validator_indices(StateData) and get_block_root_at_slot(StateData)
* switch makeAttestationData() to StateData
* sync AllTests-mainnet.md after rebase
* Revert "Revert "Upgrade database schema" (#2570)"
This reverts commit 6057c2ffb4.
* ssz: fix loading empty lists into existing instances
Not a problem earlier because we didn't reuse instances
* bump nim-eth
* bump nim-web3
The `kvstore` design we're using now turns out to not be the best way to
use `sqlite` - in particular, there are some significant benefits to
using rowid in certain situations and to keep data in separate tables.
With this branch, there are massive improvements in startup time
(seconds instead of minutes) and state/block storage and pruning times
(milliseconds instead of seconds) - these improvements can in particular
be seen on slow drives and translate directly into better attestation
performance.
* update kvstore to new keyspace design
* remove `DirStoreRef` and the hidden `--state-db-kind` option - this
was an experiment to store large blobs in files, but with the new
kvstore, there's no compelling reason to do so
* remove `DbMap` - unused and would need updating for new keyspace
design
* introduce separate tables for each data type (blocks, states etc)
* remove "WITHOUT ROWID" pessimization for tables with large blobs
* close DbSeq statements explicitly (and earlier)
* store beacon block summaries in separate table, without SSZ
compression and load them all with single query on startup
* stop storing backwards compat full states
* mark genesis beacon block as trusted
* avoid faststreams when loading SSZ data
* remove `DisagreementBehavior` (unused)
* avoid creating indexed attestation just to check signatures - above
all, don't create it when not checking signatures ;)
* avoid pointer op when adding attestation to pool
* better iterator for yielding attestations
* add metric / log for attestation packing time
* Reset cached indices when resetting cache on SSZ read
When deserializing into an existing structure, the cache should be
cleared - goes for json also. Also improve error messages.
* use IntSet rather than HashSet[ValidatorIndex]
* add bounds check before uint64 -> int conversion
* use intsets in block transitions
* remove superfluous Nim issue explanation/reference
* performance fixes
* don't mark tree cache as dirty on read-only List accesses
* store only blob in memory for keys and signatures, parse blob lazily
* compare public keys by blob instead of parsing / converting to raw
* compare Eth2Digest using non-constant-time comparison
* avoid some unnecessary validator copying
This branch will in particular speed up deposit processing which has
been slowing down block replay.
Pre (mainnet, 1600 blocks):
```
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
3450.269, 0.000, 3450.269, 3450.269, 1, Initialize DB
0.417, 0.822, 0.036, 21.098, 1400, Load block from database
16.521, 0.000, 16.521, 16.521, 1, Load state from database
27.906, 50.846, 8.104, 1507.633, 1350, Apply block
52.617, 37.029, 20.640, 135.938, 50, Apply epoch block
```
Post:
```
3502.715, 0.000, 3502.715, 3502.715, 1, Initialize DB
0.080, 0.560, 0.035, 21.015, 1400, Load block from database
17.595, 0.000, 17.595, 17.595, 1, Load state from database
15.706, 11.028, 8.300, 107.537, 1350, Apply block
33.217, 12.622, 17.331, 60.580, 50, Apply epoch block
```
* more perf fixes
* load EpochRef cache into StateCache more aggressively
* point out security concern with public key cache
* reuse proposer index from state when processing block
* avoid genericAssign in a few more places
* don't parse key when signature is unparseable
* fix `==` overload for Eth2Digest
* preallocate validator list when getting active validators
* speed up proposer index calculation a little bit
* reuse cache when replaying blocks in ncli_db
* avoid a few more copying loops
```
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
3279.158, 0.000, 3279.158, 3279.158, 1, Initialize DB
0.072, 0.357, 0.035, 13.400, 1400, Load block from database
17.295, 0.000, 17.295, 17.295, 1, Load state from database
5.918, 9.896, 0.198, 98.028, 1350, Apply block
15.888, 10.951, 7.902, 39.535, 50, Apply epoch block
0.000, 0.000, 0.000, 0.000, 0, Database block store
```
* clear full balance cache before processing rewards and penalties
```
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
3947.901, 0.000, 3947.901, 3947.901, 1, Initialize DB
0.124, 0.506, 0.026, 202.370, 363345, Load block from database
97.614, 0.000, 97.614, 97.614, 1, Load state from database
0.186, 0.188, 0.012, 99.561, 357262, Advance slot, non-epoch
14.161, 5.966, 1.099, 395.511, 11524, Advance slot, epoch
1.372, 4.170, 0.017, 276.401, 363345, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database block store
```
Validators exiting is normal, no need to scream about it
* avoid reallocating seq on big exit queue
* avoid fetching state cache when updating head (it's rarely needed)
* remove incorrectly implemented live validator counts (avoids memory
allocs)
Calculating rewards/penalties is slow due to how we compute sets of
attestations validators then use the sets for inclusion checks, to see
who attested. The dominant function during validated block processing /
epoch processing is hash set building and lookup.
This PR inverts the flow by removing the sets and creating a single
large validator status list, then applying all relevant state
attestations, then updating rewards and penalties.
This provides a 10x speedup to epoch processing which in turn speeds up
both empty slot and block processing - for example, on startup, we
replay all non-finalized blocks to prime fork choice - the same when
validating attestations or replaying states on reorg.
* misc memory and perf fixes
* use EpochRef for attestation aggregation
* compress effective balances in memory (medalla unfinalized: 4gb ->
1gb)
* avoid hitting db when rewinding to head or clearance state
* avoid hitting db when blocks can be applied to in-memory state -
speeds up startup considerably
* avoid storing epochref in fork choice
* simplify and speed up beacon block creation flow - avoids state reload
thanks to head rewind optimization
* iterator-based committee and attestation participation help avoid lots
of small memory allocations throughout epoch transition (40% speedup on
epoch processing, for example during startup)
* add constant for threshold
* evaluate block attestations under the epochref of the block - this is
what the state transition function does
* avoid copying attestation seq unnecessarily
* avoid unnecessary hashset for unslashed indices
* Lazy loading of crypto objects
* Try to fix incorrect field access by hiding fields but no luck. SSZ/Chronicles/macro bug?
* Fix incorrect blsValue access. was "aggregate" not "chronicles"
* Fix tests that rely on the internal BLSValue representation
* limit attestations kept in attestation pool
With fork choice updated, the attestation pool only needs to keep track
of attestations that will eventually end up in blocks - we can thus
limit the horizon of attestations that we keep more aggressively.
To get here, we expose getEpochRef which gets metadata about a
particular epochref, and make sure to populate it when a block is added
- this ensures that state rewinds during block addition are minimized.
In addition, we'll use the target root/epoch when validating
attestations - this helps minimize the number of different states that
we need to rewind to, in general.
* remove CandidateChains.justifiedState
unused
* remove BlockPools.Head object
* avoid quadratic quarantine loop
* fix
* add helper for beacon committee length (used for quickly validating
attestations)
* refactor some attestation checks to do cheap checks early
* validate attestation epoch before computing active validator set
* clean up documentation / comments
* fill state cache on demand
* avoid converting from uint64 to int, and where most feasible, int type conversion at all
* .len.uint64 -> .len64
* fix 32-bit compilation
* try keeping state_sim loop variable/bounds as int for 32-bit Azure
* len64 -> lenu64
* cache beacon committee size calculation
this fixes a bug in get_validator_churn_limit as well
* fix
* make committee counts consistently uint64
mixing feels like the worst of the two worlds
hash_tree_root was turning up when running beacon_node, turns out to be
repeated hash_tree_root invocations - this pr brings them back down to
normal.
this PR caches the root of a block in the SignedBeaconBlock object -
this has the potential downside that even invalid blocks will be hashed
(as part of deserialization) - later, one could imagine delaying this
until checks have passed
there's also some cleanup of the `cat=` logs which were applied randomly
and haphazardly, and to a large degree are duplicated by other
information in the log statements - in particular, topics fulfill the
same role
* restore EpochRef and flush statecaches on epoch transitions
* more targeted cache invalidation
* remove get_empty_per_epoch_cache(); implement simpler but still faster get_beacon_proposer_index()/compute_proposer_index() approach; add some abstraction layer for accessing the shuffled validator indices cache
* reduce integer type conversions
* remove most of rest of integer type conversion in compute_proposer_index()
This also assigns precise types to the constants in the minimal
and mainnet presets in order to reduce the chance of compilation
errors when custom presets are used (previously, only the custom
presets have precisely assigned types for the constants).
* update most of the remaining non-fork-choice spec refs, updating code where necessary
* revert presumably harmless compute_signing_root() change, but this way, keep things really unchanged outside inspector
* Dual headed fork choice
* fix finalizedEpoch not moving
* reduce fork choice verbosity
* Add failing tests due to pruning
* Properly handle duplicate blocks in sync
* test_block_pool also add a test for duplicate blocks
* comments addressing review
* Fix fork choice v2, was missing integrating block proposed
* remove a spurious debug writeStackTrace
* update block_sim
* Use OrderedTable to ensure that we always load parents before children in fork choice
* Load the DAG data in fork choice at init if there is some (can sync witti)
* Cluster of quarantined blocks were not properly added to the fork choice
* Workaround async gcsafe warnings
* Update blockpoool tests
* Do the callback before clearing the quarantine
* Revert OrderedTable, implement topological sort of DAG, allow forkChoice to be initialized from arbitrary finalized heads
* Make it work with latest devel - Altona readyness
* Add a recovery mechanism when forkchoice desyncs with blockpool
* add the current problematic node to the stack
* Fix rebase indentation bug (but still producing invalid block)
* Fix cache at epoch boundaries and lateBlock addition
* adopt Result[void, string] in place of some bool return signatures
* string -> cstring to reduce memory allocations; ensure all err() strings are constants, with contextual information from higher-level callers
* logScope usage fixes
* homogenize err() reporting convention
* invalid signature in deposit isn't an error
* cleanups
* fix ncli state root check flag
* add block dump to ncli_db
* limit ncli_db benchmark length
* tone down finalization logs
* introduce trusted blocks
We only store blocks whose signature we've verified in the database - as
such, there's no need to check it again, and most importantly, no need
to deserialize the signature when loading from database.
50x startup time improvement, 200x block load time improvement.
* fix rewinding when deposits have invalid signature
* speed up ancestor iteration by avoiding copy
* avoid deserializing signatures for trusted data
* load blocks lazily when rewinding (less memory used)
* chronicles workarounds
* document trustedbeaconblock
* collect signature production and verificaiton in one place
Signatures are made over data and domain - here we collect all such
activities in one place.
Also:
* security: fix cast-before-range-check
* log block/attestation verification consistently
* run block verification based on `getProposer` in its own history
* clean up some unused stuff
* import
* missing raises
* random fixes
* create dump dir on startup
* don't crash on failure to write dump
* fix a few `uint64` instances being used when indexing arrays - this
should be a compile error but isn't due to compiler bugs
* fix standalone test_block_pool compilation
* add signed block processing in ncli
* reuse cache entry instead of allocating a new one
* allow for small clock disparities when validating blocks
* reorder ssz
* split into hash_trees and ssz_serialization, roughly, for hashing and
IO
* move bitseqs into ssz (from stew)
* clean up imports
* docs, imports
tersec
Jacek Sieka
Zahary Karadjov
Mamy Ratsimbazafy