is a command-line frontend to transactional consistency checkers for black-box
databases.
In comparison to Jepsen library it is standalone and language-agnostic tool. You
can use it with tests written in any programming language and everywhere where
JVM is available. Under the hood elle-cli
uses libraries
Elle,
Knossos and
Jepsen and provides the same correctness
guarantees.
Jepsen, Elle and Knossos supports histories only in EDN (Extensible Data Notation), it is a format for serializing data that was invented by Rich Hickey, author of Clojure, for using in Clojure applications. Typical data serialized to EDN looks quite similar to JSON:
{:type :invoke, :f :read, :process 2, :time 53137939465, :index 0}
{:type :invoke, :f :transfer, :process 3, :time 53137939133, :index 1, :value {:from 5, :to 2, :amount 2}}
{:type :invoke, :f :read, :process 1, :time 53139785248, :index 2}
{:type :invoke, :f :transfer, :process 0, :time 53139856763, :index 3, :value {:from 7, :to 9, :amount 4}}
{:type :invoke, :f :read, :process 4, :time 53155597745, :index 4}
However, outside of the Clojure ecosystem EDN format is practically not used.
elle-cli
operates with operations history both in EDN and JSON (JavaScript
Object Notation) formats and can be successfully used
with histories produced by Jepsen tests in EDN format as well as with other
Jepsen-similar frameworks that produce histories in JSON format.
If you have a file with history written in EDN or JSON format, either as a
series of operation maps, or as a single vector or list containing those
operations, you can ask elle-cli
to check it for you at the command line like
so:
$ git clone https://github.com/ligurio/elle-cli
$ cd elle-cli
$ lein deps
$ lein uberjar
Compiling elle_cli.cli
Created /home/sergeyb/sources/elle-cli/target/elle-cli-0.1.8.jar
Created /home/sergeyb/sources/elle-cli/target/elle-cli-0.1.8-standalone.jar
$ java -jar target/elle-cli-0.1.8-standalone.jar --model rw-register histories/elle/rw-register.json
histories/elle/rw-register.edn true
elle-cli
converts files with histories in JSON format automatically to
Clojure data structures and prints out the names of all files you asked it to
check, followed by a tab, and then whether the history was valid. There are
three validity states:
true
means the history was validfalse
means the history was invalid:unknown
means checker was unable to complete the analysis; e.g. it ran out of memory.
In some cases conversion of history from JSON format to Clojure data structures
may fail and it is definitely a bug that should be reported. To workaround I
recommend to use a tool jet, it is a CLI to
transform between JSON and EDN, and then pass file in EDN format to elle-cli
.
An Elle's checker for write-read registers. Options are:
- consistency-models - a collection of consistency models we expect this
history to obey. Defaults to
strict-serializable
. Possible values are:consistent-view
,conflict-serializable
,cursor-stability
,forward-consistent-view
,monotonic-snapshot-read
,monotonic-view
,read-committed
,read-uncommitted
,repeatable-read
,serializable
,snapshot-isolation
,strict-serializable
,strong-serializable
,update-serializable
. - anomalies - a collection of specific anomalies you'd like to look for.
Defaults to
G0
. Possible values are:G0
,G0-process
,G0-realtime
,G1a
,G1b
,G1c
,G1c-process
,G-single
,G-single-process
,G-single-realtime
,G-nonadjacent
,G-nonadjacent-process
,G-nonadjacent-realtime
,G2-item
,G2-item-process
,G2-item-realtime
,G2-process
,GSIa
,GSIb
,incompatible-order
,dirty-update
,lost-update
,write-skew
. - cycle-search-timeout - how many milliseconds are we willing to search a
single SCC for a cycle? Default value is
1000
. - directory - where to output files, if desired. Default value is
nil
. - plot-format - either
png
orsvg
. Default value issvg
. - plot-timeout - how many milliseconds will we wait to render a SCC plot?
Default value is
5000
. - max-plot-bytes - maximum size of a cycle graph (in bytes of DOT) which
we're willing to try and render. Default value is
65536
.
Example of history:
{:type :invoke, :f :txn :value [[:w :x 2]], :process 0, :index 1}
{:type :ok, :f :txn :value [[:w :x 2]], :process 0, :index 2}
{:type :invoke, :f :txn :value [[:r :x nil]], :process 0, :index 3}
{:type :ok, :f :txn :value [[:r :x 3]], :process 0, :index 4}
{:type :invoke, :f :txn :value [[:r :x nil]], :process 0, :index 5}
{:type :ok, :f :txn :value [[:r :x 2]], :process 0, :index 6}
An Elle's checker for append and read histories. It checks for dependency cycles in append/read transactions.
The append test models the database as a collection of named lists, and performs transactions comprised of read and append operations. A read returns the value of a particular list, and an append adds a single unique element to the end of a particular list. We derive ordering dependencies between these transactions, and search for cycles in that dependency graph to identify consistency anomalies.
In terms of Elle values in operation are lists of integers. Each operation performs a transaction, comprised of micro-operations which are either reads of some value (returning the entire list) or appends (adding a single number to whatever the present value of the given list is). We detect cycles in these transactions using Elle's cycle-detection system.
Options are:
- consistency-models - a collection of consistency models we expect this
history to obey. Defaults to
strict-serializable
. Possible values are:consistent-view
,conflict-serializable
,cursor-stability
,forward-consistent-view
,monotonic-snapshot-read
,monotonic-view
,read-committed
,read-uncommitted
,repeatable-read
,serializable
,snapshot-isolation
,strict-serializable
,strong-serializable
,update-serializable
. - anomalies - a collection of specific anomalies you'd like to look for.
Defaults to
G0
. Possible values are:G0
,G0-process
,G0-realtime
,G1a
,G1b
,G1c
,G1c-process
,G-single
,G-single-process
,G-single-realtime
,G-nonadjacent
,G-nonadjacent-process
,G-nonadjacent-realtime
,G2-item
,G2-item-process
,G2-item-realtime
,G2-process
,GSIa
,GSIb
,incompatible-order
,dirty-update
,lost-update
,write-skew
. - cycle-search-timeout - how many milliseconds are we willing to search a
single SCC for a cycle? Default value is
1000
. - directory - where to output files, if desired. Default value is
nil
. - plot-format - either
png
orsvg
. Default value issvg
. - plot-timeout - how many milliseconds will we wait to render a SCC plot?
Default value is
5000
. - max-plot-bytes - maximum size of a cycle graph (in bytes of DOT) which
we're willing to try and render. Default value is
65536
.
Example of history:
{:index 2 :type :invoke, :value [[:append 255 8] [:r 253 nil]]}
{:index 3 :type :ok, :value [[:append 255 8] [:r 253 [1 3 4]]]}
{:index 4 :type :invoke, :value [[:append 256 4] [:r 255 nil] [:r 256 nil] [:r 253 nil]]}
{:index 5 :type :ok, :value [[:append 256 4] [:r 255 [2 3 4 5 8]] [:r 256 [1 2 4]] [:r 253 [1 3 4]]]}
{:index 6 :type :invoke, :value [[:append 250 10] [:r 253 nil] [:r 255 nil] [:append 256 3]]}
{:index 7 :type :ok :value [[:append 250 10] [:r 253 [1 3 4]] [:r 255 [2 3 4 5]] [:append 256 3]]}
A Jepsen's checker for bank histories.
Test simulates a set of bank accounts, one per row, and transfers money
between them at random, ensuring that no account goes negative.
An option --allow-negative-balances
change this behaviour.
Under snapshot isolation, one can prove that transfers must serialize, and the sum of all
accounts is conserved. Meanwhile, read transactions select the current balance
of all accounts. Snapshot isolation ensures those reads see a consistent
snapshot, which implies the sum of accounts in any read is constant as well.
Example of history:
{:type :invoke, :f :transfer, :process 0, :time 12613722542, :index 34, :value {:from 1, :to 0, :amount 5}}
{:type :fail, :f :transfer, :process 0, :time 12686176735, :index 35, :value {:from 1, :to 0, :amount 5}}
{:type :invoke, :f :read, :process 0, :time 12686563291, :index 36}
{:type :ok, :f :read, :process 0, :time 12799165489, :index 37, :value {0 97, 1 0, 2 0, 3 0, 4 0, 5 3, 6 0, 7 0, 8 0, 9 0}}
{:type :invoke, :f :transfer, :process 0, :time 12799587097, :index 38, :value {:from 6, :to 5, :amount 3}}
{:type :fail, :f :transfer, :process 0, :time 12903632203, :index 39, :value {:from 6, :to 5, :amount 3}}
{:type :invoke, :f :read, :process 0, :time 12903998176, :index 40}
{:type :ok, :f :read, :process 0, :time 13005165731, :index 41, :value {0 97, 1 0, 2 0, 3 0, 4 0, 5 3, 6 0, 7 0, 8 0, 9 0}}
{:type :invoke, :f :read, :process 0, :time 13005675266, :index 42}
{:type :ok, :f :read, :process 0, :time 13109721155, :index 43, :value {0 97, 1 0, 2 0, 3 0, 4 0, 5 3, 6 0, 7 0, 8 0, 9 0}}
{:type :invoke, :f :read, :process 0, :time 13110070211, :index 44}
{:type :ok, :f :read, :process 0, :time 13210540811, :index 45, :value {0 97, 1 0, 2 0, 3 0, 4 0, 5 3, 6 0, 7 0, 8 0, 9 0}}
{:type :invoke, :f :read, :process 0, :time 13210921850, :index 46}
A Jepsen's checker for counter histories.
In the counter test, we create a single record with a counter field, and
execute concurrent increments and reads of that counter. We look for cases,
where the observed value is greater than the sum of all :ok
increments, and
lower than the sum of all attempted increments. Note that this counter verifier
assumes the value monotonically increases and decrements are not allowed.
Example of history:
{:type :invoke, :f :add, :value 1, :op-index 1, :process 0, :time 10474104701, :index 0}
{:type :ok, :f :add, :value 1, :op-index 1, :process 0, :time 10584742951, :index 1}
{:type :invoke, :f :add, :value 1, :op-index 2, :process 0, :time 10686291797, :index 2}
{:type :ok, :f :add, :value 1, :op-index 2, :process 0, :time 10810489852, :index 3}
{:type :invoke, :f :add, :value 1, :op-index 3, :process 0, :time 10912309790, :index 4}
{:type :ok, :f :add, :value 1, :op-index 3, :process 0, :time 11040666263, :index 5}
A Jepsen's checker for an anomaly in parallel snapshot isolation (but which is prohibited in normal snapshot isolation). In long-fork, concurrent write transactions are observed in conflicting order.
For performance reasons, some database systems implement parallel snapshot isolation, rather than standard snapshot isolation. Parallel snapshot isolation allows an anomaly prevented by standard SI: a long fork, in which non-conflicting write transactions may be visible in incompatible orders. As an example, consider four transactions over an empty initial state:
(write x 1)
(write y 1)
(read x nil) (read y 1)
(read x 1) (read y nil)
Here, we insert two records, x
and y
. In a serializable system, one record
should have been inserted before the other. However, transaction 3 observes y
inserted before x
, and transaction 4 observes x
inserted before y
. These
observations are incompatible with a total order of inserts.
To test for this behavior, we insert a sequence of unique keys, and concurrently query for small batches of those keys, hoping to observe a pair of states in which the implicit order of insertion conflicts.
Long fork is an anomaly prohibited by snapshot isolation, but allowed by the slightly weaker model parallel snapshot isolation. In a long fork, updates to independent keys become visible to reads in a way that isn't consistent with a total order of those updates. For instance:
T1: w(x, 1)
T2: w(y, 1)
T3: r(x, 1), r(y, nil)
T4: r(x, nil), r(y, 1)
Under snapshot isolation, T1 and T2 may execute concurrently, because their write sets don't intersect. However, every transaction should observe a snapshot consistent with applying those writes in some order. Here, T3 implies T1 happened before T2, but T4 implies the opposite. We run an n-key generalization of these transactions continuously in our long fork test, and look for cases where some keys are updated out of order.
In snapshot isolated systems, reads should observe a state consistent with a total order of transactions. A long fork anomaly occurs when a pair of reads observes contradictory orders of events on distinct records - for instance, T1 observing record x before record y was created, and T2 observing y before x. In the long fork test, we insert unique rows into a table, and query small groups of those rows, looking for cases where two reads observe incompatible orders.
A Jepsen's checker for a set histories.
Given a set of :add
operations, that inserts a sequence of unique records
into a table, followed by a final :read
, that concurrently attempts to read
all of those records back, verifies that every successfully added element is
present in the read, and that the read contains only elements for which an :add
was attempted. We measure how long it takes for a record to become
durably visible, or, if it lost, how long it takes to disappear.
A linearizable set
should make every inserted element
immediately visible.
Example of history:
{:type :invoke, :f :add, :value [0 0], :process 0, :time 10529279413, :index 0}
{:type :ok, :f :add, :value [0 0], :process 0, :time 10661777878, :index 1}
{:type :invoke, :f :add, :value [0 1], :process 0, :time 10761664977, :index 2}
{:type :ok, :f :add, :value [0 1], :process 0, :time 10888511828, :index 3}
{:type :invoke, :f :add, :value [0 2], :process 0, :time 11077906807, :index 4}
{:type :ok, :f :add, :value [0 2], :process 0, :time 11209256522, :index 5}
Beware, set
and set-full
have a different computational
complexities. The set-full
checker is a lot more expensive, but
gives you precise bounds on latencies and stability of records
over time, whereas set
assumes a single read at the end of the
test.
A Jepsen's checker for a set histories. It is a more rigorous set analysis. We
allow :add
operations which add a single element, and :read
which return
all elements present at that time.
{:type :invoke, :f :add, :value [0 0], :process 0, :time 10529279413, :index 0}
{:type :ok, :f :add, :value [0 0], :process 0, :time 10661777878, :index 1}
{:type :invoke, :f :add, :value [0 1], :process 0, :time 10761664977, :index 2}
{:type :ok, :f :add, :value [0 1], :process 0, :time 10888511828, :index 3}
{:type :invoke, :f :add, :value [0 2], :process 0, :time 11077906807, :index 4}
{:type :ok, :f :add, :value [0 2], :process 0, :time 11209256522, :index 5}
{:type :invoke, :f :add, :value [0 3], :process 0, :time 11330024782, :index 6}
{:type :ok, :f :add, :value [0 3], :process 0, :time 11457989603, :index 7}
{:type :invoke, :f :add, :value [0 4], :process 0, :time 11620593669, :index 8}
{:type :ok, :f :add, :value [0 4], :process 0, :time 11745589449, :index 9}
{:type :invoke, :f :add, :value [0 5], :process 0, :time 11786251931, :index 10}
Beware, set
and set-full
have a different computational
complexities. The set-full
checker is a lot more expensive, but
gives you precise bounds on latencies and stability of records
over time, whereas set
assumes a single read at the end of the
test.
A Knossos checker for CAS (Compare-And-Set) registers. By default
competition/analysis
algorithm is used.
Example of history:
{:process 7, :type :invoke, :f :cas, :value [2 3]}
{:process 7, :type :fail, :f :cas, :value [2 3]}
{:process 8, :type :invoke, :f :write, :value 2}
{:process 8, :type :ok, :f :write, :value 2}
{:process 1, :type :invoke, :f :read, :value nil}
{:process 1, :type :ok, :f :read, :value 2}
{:process 4, :type :invoke, :f :read, :value nil}
{:process 4, :type :ok, :f :read, :value 2}
A Knossos checker for a mutex histories. Applicable to a test with single mutex
responding to :acquire
and :release
messages. By default
competition/analysis
algorithm is used.
Example of history:
{:type :invoke, :f :release, :process 1, :time 341187643, :index 0}
{:type :fail, :f :release, :process 1, :time 342667129, :error :not-held, :index 1}
{:type :invoke, :f :acquire, :process 3, :time 371408519, :index 2}
{:type :invoke, :f :release, :process 4, :time 584312016, :index 3}
{:type :fail, :f :release, :process 4, :time 585400396, :error :not-held, :index 4}
{:type :invoke, :f :release, :process 0, :time 584353142, :index 5}
{:type :fail, :f :release, :process 0, :time 585436373, :error :not-held, :index 6}
{:type :invoke, :f :release, :process 1, :time 584300961, :index 7}
{:type :fail, :f :release, :process 1, :time 585478186, :error :not-held, :index 8}
{:type :invoke, :f :acquire, :process 2, :time 584335820, :index 9}
{:type :invoke, :f :release, :process 0, :time 679093895, :index 10}
A custom checker for a comments histories.
Imagine an application which has a sequential stream of comments. Users make
comments by inserting new rows into a table. Because each request is
load-balanced to a different server, two transactions from the same user may
execute on different nodes. Now imagine that a user makes a comment C1
in
transaction T1
. T1
completes successfully. The user then realizes they made
a mistake, and posts a correction comment C2
, in transaction T2
. Meanwhile,
someone attempts to read all comments in a third transaction T3
, concurrent
with both T1
and T2
.
Example of history:
{:index 1, :type :invoke :f :read :value nil}
{:index 2, :type :invoke :f :write :value 425}
{:index 3, :type :ok :f :write :value 425}
{:index 4, :type :invoke :f :write :value 430}
{:index 5, :type :ok :f :write :value 430}
{:index 6, :type :ok :f :read :value #{2 10 15 20 34 35 38 42 43 47 51 53 59 61 71 72 82 88 89
113 119 123 129 132 145 146 163 167 176 206 216 224 230
238 243 244 255 260 292 294 299 312 316 324 325 327 330
350 356 359 360 361 363 366 367 371 376 403 410 419 422
430}}
A standalone checker for a sequential consistency, it checks that the effective
order of transactions should be consistent with the order on every client. Any
execution is the same as if all read
and write
ops were executed in some
global ordering, and the ops of each client process appear in the order
specified by its program. If a process order enforces that x
must be visible
before y
, we should always read both or neither.
To verify this, we have a single client perform a sequence of independent
transactions, inserting k1
, k2
, ..., kn
into different tables.
Concurrently, a different client attempts to read each of kn
, ..., k2
, k1
in turn. Because all inserts occur from the same process, they must also be
visible to any single process in that order. This implies that once a process
observes kn
, any subsequent read must see kn − 1
, and by induction, all
smaller keys.
Like G2 and the bank tests, this test does not verify consistency in general.
Copyright © 2021-2024 Sergey Bronnikov
Distributed under the Eclipse Public License either version 1.0 or (at your option) any later version.