Иван Бурмистров "Строго ориентированная последовательность временных событий". Выступление на Cassandra Сonf 2013
0 likes1,615 views
Ivan Burmistrov presents an algorithm for using Cassandra as an event logger. The algorithm aims to minimize time lag between writes and reads while ensuring write safety. Events are initially written in a "bad" state and only transition to "good" once no conflicting events exist. This allows reads to utilize caching while avoiding reading unstable data. The algorithm provides advantages like small time lags, cacheability and independence from synchronization, though writes are slower. Potential improvements and applications are also discussed.
![Using Cassandra for event logging
EventLogger
interface IEventLogger
{
void Write ( Event [] events ) ;
Event [] Read ( Event exclusiveEvent , int count ) ;
}
Availability for Read and Write.
Fast Write.
Read with minimal time lag.
No events in past while we making Read.
Same events chronology in any thread.
For any thread events orders for write and read are equals.](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-4-320.jpg)
![Using Cassandra for event logging
EventLogger
interface IEventLogger
{
void Write ( Event [] events ) ;
Event [] Read ( Event exclusiveEvent , int count ) ;
}
Availability for Read and Write.
Fast Write.
Read with minimal time lag.
No events in past while we making Read.
Same events chronology in any thread.
For any thread events orders for write and read are equals.](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-5-320.jpg)
![Using Cassandra for event logging
EventLogger
interface IEventLogger
{
void Write ( Event [] events ) ;
Event [] Read ( Event exclusiveEvent , int count ) ;
}
Availability for Read and Write.
Fast Write.
Read with minimal time lag.
No events in past while we making Read.
Same events chronology in any thread.
For any thread events orders for write and read are equals.](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-6-320.jpg)
![Using Cassandra for event logging
EventLogger
interface IEventLogger
{
void Write ( Event [] events ) ;
Event [] Read ( Event exclusiveEvent , int count ) ;
}
Availability for Read and Write.
Fast Write.
Read with minimal time lag.
No events in past while we making Read.
Same events chronology in any thread.
For any thread events orders for write and read are equals.](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-7-320.jpg)
![Using Cassandra for event logging
EventLogger
interface IEventLogger
{
void Write ( Event [] events ) ;
Event [] Read ( Event exclusiveEvent , int count ) ;
}
Availability for Read and Write.
Fast Write.
Read with minimal time lag.
No events in past while we making Read.
Same events chronology in any thread.
For any thread events orders for write and read are equals.](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-8-320.jpg)
![Using Cassandra for event logging
EventLogger
interface IEventLogger
{
void Write ( Event [] events ) ;
Event [] Read ( Event exclusiveEvent , int count ) ;
}
Availability for Read and Write.
Fast Write.
Read with minimal time lag.
No events in past while we making Read.
Same events chronology in any thread.
For any thread events orders for write and read are equals.](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-9-320.jpg)
![Using Cassandra for event logging
EventLogger
interface IEventLogger
{
void Write ( Event [] events ) ;
Event [] Read ( Event exclusiveEvent , int count ) ;
}
Availability for Read and Write.
Fast Write.
Read with minimal time lag.
No events in past while we making Read.
Same events chronology in any thread.
For any thread events orders for write and read are equals.](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-10-320.jpg)
![Using Cassandra for event logging
Obvious solution
Event
string EventId;
long Timestamp;
byte[] Content;](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-11-320.jpg)
![Using Cassandra for event logging
Obvious solution
Write(Event[] events)
Write columns for each event e:
e.Timestamp = NowTicks;
RowKey = e.Timestamp / Partition;
ColumnName = e.Timestamp + ":" + e.EventId;
ColumnValue = e.Content;
Read(Event exclusiveEvent, int count)
Execute get_slice from exclusive column;
Skip events with Timestamp > NowTicks - TimeLag.](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-12-320.jpg)
![Using Cassandra for event logging
Obvious solution
Write(Event[] events)
Write columns for each event e:
e.Timestamp = NowTicks;
RowKey = e.Timestamp / Partition;
ColumnName = e.Timestamp + ":" + e.EventId;
ColumnValue = e.Content;
Read(Event exclusiveEvent, int count)
Execute get_slice from exclusive column;
Skip events with Timestamp > NowTicks - TimeLag.](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-13-320.jpg)
![Using Cassandra for event logging
Our algorithm
Event
string EventId;
long Timestamp;
byte[] Content;
State State (Good or Bad);](https://image.slidesharecdn.com/random-131213031018-phpapp01/85/Cassandra-onf-2013-21-320.jpg)
Иван Бурмистров "Строго ориентированная последовательность временных событий". Выступление на Cassandra Сonf 2013
- 1. Using Cassandra for event logging Presentation Using Cassandra for event logging Ivan Burmistrov SKB Kontur December 9th, 2013
- 2. Using Cassandra for event logging EDI We develop EDI (Electronic data interchange) system.
- 3. Using Cassandra for event logging EDI We develop EDI (Electronic data interchange) system.
- 4. Using Cassandra for event logging EventLogger interface IEventLogger { void Write ( Event [] events ) ; Event [] Read ( Event exclusiveEvent , int count ) ; } Availability for Read and Write. Fast Write. Read with minimal time lag. No events in past while we making Read. Same events chronology in any thread. For any thread events orders for write and read are equals.
- 5. Using Cassandra for event logging EventLogger interface IEventLogger { void Write ( Event [] events ) ; Event [] Read ( Event exclusiveEvent , int count ) ; } Availability for Read and Write. Fast Write. Read with minimal time lag. No events in past while we making Read. Same events chronology in any thread. For any thread events orders for write and read are equals.
- 6. Using Cassandra for event logging EventLogger interface IEventLogger { void Write ( Event [] events ) ; Event [] Read ( Event exclusiveEvent , int count ) ; } Availability for Read and Write. Fast Write. Read with minimal time lag. No events in past while we making Read. Same events chronology in any thread. For any thread events orders for write and read are equals.
- 7. Using Cassandra for event logging EventLogger interface IEventLogger { void Write ( Event [] events ) ; Event [] Read ( Event exclusiveEvent , int count ) ; } Availability for Read and Write. Fast Write. Read with minimal time lag. No events in past while we making Read. Same events chronology in any thread. For any thread events orders for write and read are equals.
- 8. Using Cassandra for event logging EventLogger interface IEventLogger { void Write ( Event [] events ) ; Event [] Read ( Event exclusiveEvent , int count ) ; } Availability for Read and Write. Fast Write. Read with minimal time lag. No events in past while we making Read. Same events chronology in any thread. For any thread events orders for write and read are equals.
- 9. Using Cassandra for event logging EventLogger interface IEventLogger { void Write ( Event [] events ) ; Event [] Read ( Event exclusiveEvent , int count ) ; } Availability for Read and Write. Fast Write. Read with minimal time lag. No events in past while we making Read. Same events chronology in any thread. For any thread events orders for write and read are equals.
- 10. Using Cassandra for event logging EventLogger interface IEventLogger { void Write ( Event [] events ) ; Event [] Read ( Event exclusiveEvent , int count ) ; } Availability for Read and Write. Fast Write. Read with minimal time lag. No events in past while we making Read. Same events chronology in any thread. For any thread events orders for write and read are equals.
- 11. Using Cassandra for event logging Obvious solution Event string EventId; long Timestamp; byte[] Content;
- 12. Using Cassandra for event logging Obvious solution Write(Event[] events) Write columns for each event e: e.Timestamp = NowTicks; RowKey = e.Timestamp / Partition; ColumnName = e.Timestamp + ":" + e.EventId; ColumnValue = e.Content; Read(Event exclusiveEvent, int count) Execute get_slice from exclusive column; Skip events with Timestamp > NowTicks - TimeLag.
- 13. Using Cassandra for event logging Obvious solution Write(Event[] events) Write columns for each event e: e.Timestamp = NowTicks; RowKey = e.Timestamp / Partition; ColumnName = e.Timestamp + ":" + e.EventId; ColumnValue = e.Content; Read(Event exclusiveEvent, int count) Execute get_slice from exclusive column; Skip events with Timestamp > NowTicks - TimeLag.
- 14. Using Cassandra for event logging Advantages and disadvantages Advantages Easy implementation. Write action is very fast. Disadvantages TimeLag = TimeLag > TimeLag > TimeLag > ? WriteTimeout + SyncDiff; ReadTimeout + SyncDiff; Cassandra worst write time; We can’t cache events in unstable zone, if we want to read events from it. I.e. more requests −→ more CPU for cassandra.
- 15. Using Cassandra for event logging Advantages and disadvantages Advantages Easy implementation. Write action is very fast. Disadvantages TimeLag = TimeLag > TimeLag > TimeLag > ? WriteTimeout + SyncDiff; ReadTimeout + SyncDiff; Cassandra worst write time; We can’t cache events in unstable zone, if we want to read events from it. I.e. more requests −→ more CPU for cassandra.
- 16. Using Cassandra for event logging Advantages and disadvantages Advantages Easy implementation. Write action is very fast. Disadvantages TimeLag = TimeLag > TimeLag > TimeLag > ? WriteTimeout + SyncDiff; ReadTimeout + SyncDiff; Cassandra worst write time; We can’t cache events in unstable zone, if we want to read events from it. I.e. more requests −→ more CPU for cassandra.
- 17. Using Cassandra for event logging Advantages and disadvantages Advantages Easy implementation. Write action is very fast. Disadvantages TimeLag = TimeLag > TimeLag > TimeLag > ? WriteTimeout + SyncDiff; ReadTimeout + SyncDiff; Cassandra worst write time; We can’t cache events in unstable zone, if we want to read events from it. I.e. more requests −→ more CPU for cassandra.
- 18. Using Cassandra for event logging Advantages and disadvantages Advantages Easy implementation. Write action is very fast. Disadvantages TimeLag = TimeLag > TimeLag > TimeLag > ? WriteTimeout + SyncDiff; ReadTimeout + SyncDiff; Cassandra worst write time; We can’t cache events in unstable zone, if we want to read events from it. I.e. more requests −→ more CPU for cassandra.
- 19. Using Cassandra for event logging Advantages and disadvantages Advantages Easy implementation. Write action is very fast. Disadvantages TimeLag = TimeLag > TimeLag > TimeLag > ? WriteTimeout + SyncDiff; ReadTimeout + SyncDiff; Cassandra worst write time; We can’t cache events in unstable zone, if we want to read events from it. I.e. more requests −→ more CPU for cassandra.
- 20. Using Cassandra for event logging Advantages and disadvantages Advantages Easy implementation. Write action is very fast. Disadvantages TimeLag = TimeLag > TimeLag > TimeLag > ? WriteTimeout + SyncDiff; ReadTimeout + SyncDiff; Cassandra worst write time; We can’t cache events in unstable zone, if we want to read events from it. I.e. more requests −→ more CPU for cassandra.
- 21. Using Cassandra for event logging Our algorithm Event string EventId; long Timestamp; byte[] Content; State State (Good or Bad);
- 22. Using Cassandra for event logging Our algorithm Write(Event event) Timestamp = NowTicks, State = Bad, write event. Are there Good events with bigger Timestamp? No? State = Good, rewrite event. Yes? Delete event and try again.
- 23. Using Cassandra for event logging Our algorithm Write(Event event) Timestamp = NowTicks, State = Bad, write event. Are there Good events with bigger Timestamp? No? State = Good, rewrite event. Yes? Delete event and try again.
- 24. Using Cassandra for event logging Our algorithm Write(Event event) Timestamp = NowTicks, State = Bad, write event. Are there Good events with bigger Timestamp? No? State = Good, rewrite event. Yes? Delete event and try again.
- 25. Using Cassandra for event logging Our algorithm Write(Event event) Timestamp = NowTicks, State = Bad, write event. Are there Good events with bigger Timestamp? No? State = Good, rewrite event. Yes? Delete event and try again.
- 26. Using Cassandra for event logging Our algorithm Read We can read events (same as obvious solution) while it’s Good.
- 27. Using Cassandra for event logging Our algorithm Read We can read events (same as obvious solution) while it’s Good.
- 28. Using Cassandra for event logging Our algorithm TimeLag = ? TimeLag do not depends on timeouts. Less write intensity −→ less TimeLag. Less cassandra write time −→ less TimeLag.
- 29. Using Cassandra for event logging Our algorithm TimeLag = ? TimeLag do not depends on timeouts. Less write intensity −→ less TimeLag. Less cassandra write time −→ less TimeLag.
- 30. Using Cassandra for event logging Our algorithm TimeLag = ? TimeLag do not depends on timeouts. Less write intensity −→ less TimeLag. Less cassandra write time −→ less TimeLag.
- 31. Using Cassandra for event logging Improvements Improvements "Are there Good events with bigger Timestamp?" for O(1). Queue write requests and process its by batches. We could remove the dependency from time sync, if NowTicks will be monotonous.
- 32. Using Cassandra for event logging Improvements Improvements "Are there Good events with bigger Timestamp?" for O(1). Queue write requests and process its by batches. We could remove the dependency from time sync, if NowTicks will be monotonous.
- 33. Using Cassandra for event logging Improvements Improvements "Are there Good events with bigger Timestamp?" for O(1). Queue write requests and process its by batches. We could remove the dependency from time sync, if NowTicks will be monotonous.
- 34. Using Cassandra for event logging Sharding Simple sharding We can divide all entities by independent parts. For each part we can use separate eventLogger.
- 35. Using Cassandra for event logging Sharding Simple sharding We can divide all entities by independent parts. For each part we can use separate eventLogger.
- 36. Using Cassandra for event logging Sharding Sharding for reading Timestamp = NowTicks, State = Bad, write events in different rows. Are there Good events with bigger Timestamp in any row? No? State = Good, rewrite events. Yes? Delete event and try again.
- 37. Using Cassandra for event logging Sharding Sharding for reading Timestamp = NowTicks, State = Bad, write events in different rows. Are there Good events with bigger Timestamp in any row? No? State = Good, rewrite events. Yes? Delete event and try again.
- 38. Using Cassandra for event logging Sharding Sharding for reading Timestamp = NowTicks, State = Bad, write events in different rows. Are there Good events with bigger Timestamp in any row? No? State = Good, rewrite events. Yes? Delete event and try again.
- 39. Using Cassandra for event logging Advantages and disadvantages Advantages Small size of time lag. For 3500 writes per sec worst time lag is 50ms and 10 events. More cacheable. Easy sharding. Can be independed from time sync. Disadvantages More slow writes. But still enough fast. Limit for writing threads.
- 40. Using Cassandra for event logging Advantages and disadvantages Advantages Small size of time lag. For 3500 writes per sec worst time lag is 50ms and 10 events. More cacheable. Easy sharding. Can be independed from time sync. Disadvantages More slow writes. But still enough fast. Limit for writing threads.
- 41. Using Cassandra for event logging Advantages and disadvantages Advantages Small size of time lag. For 3500 writes per sec worst time lag is 50ms and 10 events. More cacheable. Easy sharding. Can be independed from time sync. Disadvantages More slow writes. But still enough fast. Limit for writing threads.
- 42. Using Cassandra for event logging Advantages and disadvantages Advantages Small size of time lag. For 3500 writes per sec worst time lag is 50ms and 10 events. More cacheable. Easy sharding. Can be independed from time sync. Disadvantages More slow writes. But still enough fast. Limit for writing threads.
- 43. Using Cassandra for event logging Advantages and disadvantages Advantages Small size of time lag. For 3500 writes per sec worst time lag is 50ms and 10 events. More cacheable. Easy sharding. Can be independed from time sync. Disadvantages More slow writes. But still enough fast. Limit for writing threads.
- 44. Using Cassandra for event logging Advantages and disadvantages Advantages Small size of time lag. For 3500 writes per sec worst time lag is 50ms and 10 events. More cacheable. Easy sharding. Can be independed from time sync. Disadvantages More slow writes. But still enough fast. Limit for writing threads.
- 45. Using Cassandra for event logging Another applications Another applications Indexing. We can use event log as notification about change state of some objects −→ indexing. Cacheable is very useful here! Message queues.
- 46. Using Cassandra for event logging Another applications Another applications Indexing. We can use event log as notification about change state of some objects −→ indexing. Cacheable is very useful here! Message queues.