NEW
Font size
S
M
L
XL
WorksheetsBig Data 4: Presto
Total questions: 75
Worksheet time: 38mins
Name
Class
Date
1.
In the chapter title, PrestoDB is positioned primarily as which kind of system component?
a)
A distributed SQL query processing engine
b)
A document-oriented primary database
c)
A key-value storage engine
d)
A graph database with native traversal
e)
A log-based stream processing framework
2.
According to the history slide, which pairing best matches the core design intent of Presto?
a)
Interactive querying combined with data-warehouse-like speed and Facebook-scale scalability
b)
Batch ETL combined with strict serializable transactions
c)
Low-latency writes combined with multi-master replication
d)
High-throughput message ingestion combined with exactly-once stream semantics
e)
Schema-on-write storage combined with embedded BI dashboards
3.
The motivation slide describes a workaround used to support dashboards when Hive is not interactive. What is that workaround?
a)
Persist daily-batch results into an interactive database for quick response
b)
Replace HDFS with an interactive database as the primary storage layer
c)
Run BI tools directly against Hive by increasing reducer parallelism
d)
Export all dashboards into static reports to avoid query execution
e)
Disable ODBC/JDBC connectivity and rely on file copies to user machines
4.
The motivation slide notes that some data is not stored in HDFS. What extra step does this create for analysis?
a)
Copy the data into HDFS before analyzing it
b)
Convert the data into MapReduce jobs without loading it anywhere
c)
Rewrite the BI dashboards to read from local files on the client
d)
Store the data only in PostgreSQL and avoid HDFS entirely
e)
Use ODBC to query HDFS directly without any metadata
5.
In the depicted two-platform pipeline, which component is explicitly labeled as serving interactive queries before data reaches the dashboard?
a)
PostgreSQL, etc.
b)
HDFS
c)
Hive
d)
Daily/Hourly Batch
e)
Commercial BI Tools
6.
Based on the diagram separating a batch analysis platform from a visualization platform, which grouping best matches that separation?
a)
Batch: HDFS and Hive; Visualization: dashboards and BI tools (via an interactive DB)
b)
Batch: dashboards and BI tools; Visualization: HDFS and Hive
c)
Batch: PostgreSQL; Visualization: Hive and HDFS
d)
Batch: Commercial BI Tools; Visualization: Daily/Hourly Batch
e)
Batch: Client; Visualization: Coordinator
7.
In the annotated two-platform setup, which consequence is explicitly attributed to operating separate batch analysis and visualization platforms?
a)
More work to manage two platforms
b)
Automatic fault tolerance for interactive queries
c)
Ability to query live HDFS data directly from BI tools
d)
Elimination of extra costs from interactive storage
e)
Guaranteed linear scalability of the interactive database layer
8.
The annotated diagram highlights that the dashboard cannot query "live" data directly. In that context, what does "live" most directly refer to?
a)
Data residing in HDFS that is not materialized into the interactive database
b)
Dashboard caches stored in the browser memory
c)
Row-level locks held inside PostgreSQL
d)
SQL query text sent from the client to the coordinator
e)
Metadata stored in the Hive Metastore
9.
The transformation diagram replaces a two-platform workflow with a single interactive path. Which change enables the dashboard to query without first copying results into PostgreSQL?
a)
Introduce Presto as the interactive query layer reading from HDFS
b)
Remove Hive so that only daily/hourly batch remains
c)
Move dashboards into HDFS so they can run next to the data
d)
Replace HDFS with an interactive database as the only storage
e)
Convert all queries into MapReduce jobs executed by Hive
10.
In the multi-source architecture diagram, Presto sits between several data stores and the dashboard. What capability is the diagram emphasizing?
a)
SQL queries can span multiple underlying data sources through Presto
b)
Presto physically copies all data sources into HDFS before querying
c)
Presto replaces Hive as the daily/hourly batch ETL engine
d)
Presto requires all data to be stored in Cassandra to be queried
e)
The dashboard must connect separately to each data source
11.
The same diagram shows both Hive and Presto connected to HDFS. Which role division is most consistent with that picture?
a)
Hive for batch processing; Presto for interactive querying
b)
Hive for interactive querying; Presto for batch ETL
c)
Hive and Presto both act only as storage/metadata systems
d)
Hive is a connector plugin; Presto is a storage engine
e)
Hive runs inside the dashboard; Presto runs inside HDFS
12.
The diagram extends the interactive path beyond dashboards. Besides dashboards, which consumer is shown receiving query results from Presto?
a)
Commercial BI Tools
b)
Hive Metastore
c)
Discovery Service
d)
Daily/Hourly Batch Scheduler
e)
HDFS NameNode
13.
The "What Presto can do" slide includes a boundary between interactive querying and ETL. Which statement matches that boundary?
a)
MapReduce and Hive remain necessary for ETL even when Presto is used for interactive queries
b)
Presto fully replaces MapReduce and Hive for all ETL workloads
c)
ETL is performed by dashboards through ODBC connections
d)
Presto requires all ETL outputs to be stored in commercial BI tools
e)
ETL is avoided by storing all data in PostgreSQL only
14.
The same slide lists a plugin mechanism and querying across multiple data sources. What is the most direct implication of that pairing?
a)
Support for new data sources can be added via connectors without changing the core query engine
b)
Query planning becomes impossible without converting all sources into ORC
c)
ODBC/JDBC becomes unreliable when multiple sources are used
d)
Presto must duplicate each data source into a single storage system
e)
Only Hive can be queried because plugins are limited to HDFS
15.
The deployment slide lists scale metrics (nodes, queries per day, data processed). Which interpretation is most consistent with those metrics?
a)
Presto was operated as a large shared service handling many interactive queries at significant throughput
b)
Presto was used only for single-user offline batch jobs with long runtimes
c)
Presto was deployed as an embedded library inside dashboards only
d)
Presto replaced HDFS as the primary persistent storage layer
e)
Presto eliminated the need for multiple geographical regions
16.
The slide showing many company logos is primarily used to communicate which point about Presto?
a)
It has been adopted in a wide range of organizations and environments
b)
It can only be deployed on a single cloud provider
c)
It is limited to academic demonstration workloads
d)
It requires proprietary BI tools to function
e)
It only works when all data is stored in PostgreSQL
17.
In the architecture diagram, which component is the first point of contact for the client issuing a query?
a)
Coordinator
b)
Worker
c)
Connector Plugin
d)
Storage / Metadata
e)
Discovery Service
18.
In the same diagram, which element is shown as the interface to storage and metadata for both coordinator and workers?
a)
Connector Plugin
b)
Discovery Service
c)
Client
d)
Worker
e)
Coordinator
19.
In the step labeled "find servers in a cluster", which component is highlighted as enabling server discovery?
a)
Discovery Service
b)
Connector Plugin
c)
Hive Metastore
d)
Dashboard
e)
Commercial BI Tools
20.
The step diagram states that the client sends a query using HTTP. Which alternative protocol is explicitly not shown as the client-to-coordinator transport in that step?
a)
A disk-based batch job submission interface
b)
HTTP
c)
JSON over HTTP
d)
A client request to the coordinator
e)
A network request originating from the client
21.
In the step where the coordinator builds a query plan, what dependency is shown as necessary for planning?
a)
Metadata such as table schema provided by a connector plugin
b)
Precomputed dashboard caches stored in the client browser
c)
A completed MapReduce job for every referenced table
d)
A full materialization of all source data into PostgreSQL
e)
A replicated copy of the query results in HDFS before planning
22.
The same step explicitly labels what the connector plugin provides. Which item matches that label?
a)
Table schema (and related metadata)
b)
Final query results in CSV format
c)
Worker node list for scheduling
d)
Interactive dashboards and visual widgets
e)
User authentication tokens for HTTP requests
23.
In the step-by-step execution diagram, what action follows planning and is explicitly assigned to the coordinator?
a)
Send tasks to workers
b)
Execute all operators locally on the coordinator only
c)
Write intermediate results to disk between stages
d)
Copy source data into the Hive Metastore
e)
Return results directly without any worker involvement
24.
The execution step labeled "Workers read data" shows workers reading through what mechanism?
a)
A connector plugin
b)
The discovery service
c)
A dashboard connector
d)
A MapReduce reducer output file
e)
A client-side JDBC driver cache
25.
The step labeled "Workers run tasks" emphasizes where task execution occurs. According to the diagram, tasks run primarily where?
a)
In memory on the workers
b)
On disk on the coordinator
c)
Inside the discovery service
d)
Only inside the storage system
e)
Exclusively inside BI tools
26.
The final step indicates how results are returned. According to the diagram, the client gets the result from which component?
a)
A worker
b)
The discovery service
c)
The storage/metadata system directly
d)
The connector plugin only
e)
Hive
27.
In the architecture diagram repeated without the title, which component is positioned directly between workers and the storage/metadata system?
a)
Connector Plugin
b)
Client
c)
Coordinator
d)
Discovery Service
e)
Dashboard
28.
The connectors slide describes connectors as plugins. Which combination matches what the slide states about their implementation and purpose?
a)
Written in Java and used to access storage and metadata
b)
Written in SQL and used to store query results persistently
c)
Written in Python and used to replace the coordinator
d)
Written in C and used to implement dashboards
e)
Written in JavaScript and used to provide the HTTP protocol
29.
According to the connectors slide, which pairing correctly matches what connectors provide to coordinators versus workers?
a)
Coordinators get table schema; workers get table rows
b)
Coordinators get table rows; workers get table schema
c)
Both coordinators and workers get only the node list
d)
Both coordinators and workers get only JDBC drivers
e)
Coordinators get dashboards; workers get BI tools
30.
In the Hive connector diagram, the connector is shown interfacing with which backend systems?
a)
HDFS and the Hive Metastore
b)
Cassandra and ZooKeeper
c)
MySQL and Redis
d)
S3 and DynamoDB
e)
Dashboard and ODBC
31.
In the Cassandra connector diagram, the connector is shown interfacing with which backend system?
a)
Cassandra
b)
HDFS / Metastore
c)
Commercial BI Tools
d)
Discovery Service
e)
SQL Parser
32.
The "Multiple connectors in a query" diagram implies that a single query can do what, without first consolidating all data into one store?
a)
Access multiple data sources through different connectors within the same query
b)
Force all tables to be stored in HDFS before execution
c)
Run only against a single connector chosen at startup
d)
Materialize all joins into PostgreSQL prior to planning
e)
Disable the coordinator and run queries from workers directly
33.
In the same multi-connector diagram, which component is consistently shown distributing work to workers while connectors talk to their own backends?
a)
Coordinator
b)
Storage / Metadata
c)
Dashboard
d)
Commercial BI Tools
e)
External data sources
34.
The distributed architecture slide states that Presto is NOT a database. Which consequence follows most directly from that statement as presented?
a)
Presto provides SQL over existing data stores rather than storing the data itself
b)
Presto cannot query any external system and requires local files only
c)
Presto automatically performs daily ETL and persists results by default
d)
Presto replaces connector plugins by embedding storage engines
e)
Presto requires an exclusive proprietary client protocol
35.
The same slide notes the client protocol is HTTP + JSON and lists multiple language bindings. What usage pattern does this most strongly support?
a)
Clients in many languages can submit queries over HTTP and parse JSON responses
b)
Only Java clients can connect because JSON requires a JVM
c)
Queries must be executed via MapReduce because HTTP is batch-only
d)
Workers must be written in each client language to run tasks
e)
Connectors must be rewritten in each client language to provide schemas
36.
The "Coordinator HA" diagram shows two coordinators connected into the same cluster. Which interpretation best matches the purpose of this duplication?
a)
Improve availability by avoiding a single coordinator as a single point of failure
b)
Increase storage capacity by replicating the storage/metadata system
c)
Replace workers with coordinators for data scanning
d)
Eliminate the need for a discovery service
e)
Force all tasks to run on coordinators instead of workers
37.
In the same HA diagram, which component is still shown centrally connected to both coordinators and workers, suggesting it supports cluster membership discovery?
a)
Discovery Service
b)
Commercial BI Tools
c)
PostgreSQL
d)
Hive Metastore
e)
Dashboard
38.
The execution model slide contrasts Presto with MapReduce and states that Presto query plans are based on what structure?
a)
A DAG (directed acyclic graph)
b)
A strict linear pipeline with a single global barrier
c)
A ring topology of replicated reducers
d)
A two-phase commit transaction log
e)
A master-only sequential execution trace
39.
The same slide lists responsibilities split between coordinator and workers. Which mapping is consistent with the slide?
a)
Coordinator: SQL parser, query planner, execution planner; Workers: task execution scheduler
b)
Coordinator: table scanning; Workers: SQL parsing and query planning
c)
Coordinator: connector implementations; Workers: metadata storage
d)
Coordinator: dashboard rendering; Workers: BI tool connectivity
e)
Coordinator: HDFS storage; Workers: discovery service
40.
In the planning pipeline diagram, which inputs are shown feeding into the logical planner stage?
a)
Connector and optimizer informed by metadata such as table schema
b)
Node list from the discovery server only
c)
Final query results produced by workers
d)
Disk-based shuffle files written by reducers
e)
Dashboard configuration and visualization widgets
41.
In the same diagram, the execution planner is shown relying on which component to obtain a node list?
a)
NodeManager connected to the discovery server
b)
Optimizer connected to metadata
c)
SQL Parser connected to the client
d)
Connector connected to storage only
e)
Distributed planner connected to BI tools
42.
In the query planner example, the distributed query plan introduces both partial aggregation and final aggregation. What is the most direct reason for this split as implied by the diagram?
a)
Reduce data volume before the exchange by aggregating locally, then finalize after data transfer
b)
Ensure all aggregation happens only on the coordinator for correctness
c)
Force the engine to write intermediate results to disk between stages
d)
Avoid using a GROUP BY operator by converting it into a table scan
e)
Eliminate the need for any exchange by duplicating the full table on each worker
43.
In the same example, which operator name most directly signals inter-worker data movement in the distributed plan?
a)
Exchange
b)
Output
c)
Table scan
d)
GROUP BY
e)
Sink
44.
In the "Query Planner - Stages" diagram, which stages are explicitly annotated as involving inter-worker data transfer?
a)
Stage-0 and Stage-2
b)
Stage-1 only
c)
Stage-0 only
d)
Stage-2 only
e)
All stages equally with no distinction
45.
The same stages diagram labels Stage-1 as "pipelined aggregation". Which operator sequence is shown within that stage block?
a)
Sink, Final aggregation, Exchange
b)
Output, Exchange
c)
Sink, Partial aggregation, Table scan
d)
Hash, LeftJoin, LocalShuffle
e)
Filter, Scan
46.
The stages slide defines a stage as a part of a plan that can be executed in parallel across workers. Which statement best matches that definition?
a)
Different workers run the same computation on different input partitions within a stage
b)
Only the coordinator runs computation while workers remain idle
c)
A stage is a single SQL statement executed on a single node
d)
A stage is always a disk-backed MapReduce reduce phase
e)
A stage is a connector plugin that exposes table schema
47.
The stages slide lists shuffles as buffered in-memory transfers between stages and notes their costs. Which cost is explicitly listed?
a)
Higher CPU overhead
b)
Guaranteed elimination of network traffic
c)
Automatic persistence of intermediate results to disk
d)
Improved fault tolerance via checkpointing
e)
Reduced memory pressure by spilling by default
48.
The execution planner diagram shows a distributed plan combined with a node list (two workers) to form a concrete execution plan. What variable does the node list most directly determine?
a)
Which workers receive instances of each plan fragment for execution
b)
The SQL syntax accepted by the parser
c)
The table schemas returned by connectors
d)
Whether queries are converted into MapReduce jobs
e)
Whether results are returned via disk files
49.
The "Execution Planner - Tasks" slide states "1 task / worker / stage". If a stage is scheduled on 4 workers, how many tasks correspond to that stage?
a)
4 tasks
b)
1 task
c)
2 tasks
d)
8 tasks
e)
It must be 0 tasks until all stages finish
50.
The tasks slide states that the coordinator distributes plan stages to workers in the form of executable tasks. Which relationship between stages, tasks, and workers is explicitly stated?
a)
One task runs on one worker for one stage
b)
One stage runs on one worker for one task
c)
One task spans multiple workers within one stage
d)
One worker runs multiple stages without tasks
e)
One connector provides tasks to the coordinator
51.
The same slide states that a task may have multiple pipelines and a pipeline is a chain of operators. Which description best matches a pipeline as used here?
a)
A sequential operator chain inside a task that processes data as it flows through
b)
A replicated storage layout that determines partitioning and sorting
c)
A discovery mechanism that finds servers in a cluster
d)
A client protocol that encodes results in JSON
e)
A fault-tolerant checkpoint file written to disk between stages
52.
The scheduling slide contrasts "all-at-once" versus "phased" scheduling. Which property is explicitly associated with all-at-once scheduling?
a)
Schedules all stages concurrently so data is processed as soon as it is available
b)
Schedules only leaf stages and blocks intermediate stages until completion
c)
Minimizes memory by forcing a strict stage-by-stage barrier
d)
Writes shuffle data to disk to improve reliability
e)
Requires pre-enumerating all splits before any task can start
53.
The same scheduling slide states that phased scheduling improves memory efficiency for which use case?
a)
Batch Analytics
b)
Interactive Analytics
c)
Developer/Advertiser Analytics
d)
A/B Testing
e)
Dashboard rendering
54.
The task scheduling slide distinguishes leaf stages from intermediate stages. Which placement constraint is explicitly associated with leaf stages?
a)
Task placement considers network and connector constraints such as co-location with storage nodes
b)
Tasks can be placed on any worker node without considering data layout
c)
Tasks must always run on the coordinator for correctness
d)
Tasks must always run on the discovery service to access the node list
e)
Tasks must be executed only after all splits are enumerated up front
55.
According to the same slide, what is explicitly true about intermediate stages?
a)
Tasks can be placed on any worker node
b)
Tasks must be co-located with storage nodes
c)
Tasks must be written in Java as connectors
d)
Tasks always perform table scans and never shuffles
e)
Tasks are scheduled only after the client receives all results
56.
In the "Execution Planner - Split" diagram, which scenario is explicitly associated with "many threads per worker"?
a)
Many splits per task during a table scan
b)
One split per worker during output
c)
One split per task during final aggregation
d)
Zero splits because the coordinator reads all data
e)
One split shared across all workers via the discovery service
57.
The split scheduling slide defines a split as an opaque handle to what?
a)
An addressable chunk of data in an external storage system or intermediate result
b)
A SQL statement fragment stored in the client cache
c)
A physical table layout describing partitioning and sorting
d)
A stage boundary that enforces a global barrier
e)
A connector implementation written in Java
58.
The same slide states that splits are assigned lazily and can be assigned to the worker with the shortest queue. Which benefit is explicitly mentioned for this lazy approach?
a)
Reduced metadata memory usage on the coordinator
b)
Guaranteed fault tolerance through retrying failed splits
c)
Elimination of all network transfers between workers
d)
Automatic conversion of queries into MapReduce jobs
e)
Mandatory disk spilling for large aggregations
59.
The data layouts slide states that a table can have multiple layouts with different properties. What is the optimizer described as doing with these layouts?
a)
Choosing the best layout for a given query
b)
Using all layouts simultaneously for every query
c)
Ignoring layout properties such as partitioning and sorting
d)
Replacing connectors with layouts to access metadata
e)
Requiring layouts to include all columns and all rows
60.
The same slide suggests a tuning approach. Which action aligns with that suggestion?
a)
Add a new physical layout so the optimizer has a better option for certain queries
b)
Disable the optimizer so it cannot choose among layouts
c)
Convert all data sources into a single connector to remove layouts
d)
Force every query to scan all columns to avoid layout selection
e)
Remove partitioning and sorting to make layouts identical
61.
The predicate pushdown slide says the engine provides connectors with a two-part constraint. Which pair matches those two parts?
a)
Domain of values (ranges and nullability) and a black box predicate
b)
Node list and task list
c)
SQL parser output and AST
d)
Partitioning key and sort key only
e)
Disk spill file path and retry policy
62.
In the example showing a logical plan and a distributed plan, which operation is performed before the left join on the lineitem side in the logical plan?
a)
A filter with condition discount = 0
b)
A final aggregation of SUM(tax)
c)
An output operator emitting the final rows
d)
A local shuffle after the join
e)
A scan of the orders table
63.
In the MapReduce vs. Presto comparison, which behavior is explicitly associated with MapReduce but not with Presto?
a)
Writing data to disk between stages and waiting between stages
b)
Memory-to-memory data transfer with no disk IO
c)
All stages being pipelined with no wait time
d)
A requirement that data chunks fit in memory
e)
Executing tasks across a DAG-like plan
64.
The same comparison lists tradeoffs for Presto. Which pair of limitations is explicitly stated?
a)
No fault tolerance and data chunks must fit in memory
b)
High disk IO and strict stage barriers
c)
Requires MapReduce and cannot run interactively
d)
Only queries a single data source and cannot use connectors
e)
Must store all results in PostgreSQL before returning
65.
The query execution slide states a failure property: if one task fails, what happens?
a)
All tasks fail at once and the query fails
b)
Only that task is retried while the query continues
c)
Only the stage containing that task is skipped
d)
The worker process terminates and the cluster shuts down
e)
The coordinator converts execution into MapReduce automatically
66.
The same slide states an in-memory limitation related to aggregation. What outcome is explicitly stated when aggregated data does not fit in memory?
a)
The query fails
b)
The engine spills automatically to disk and continues
c)
The engine switches to Hive for the aggregation step
d)
The connector rewrites the query to avoid aggregation
e)
The worker is permanently removed from the cluster
67.
In the Qubole Presto benchmark chart, the y-axis is labeled as a speedup over S3-Hive. What does a larger bar indicate in this context?
a)
Higher speedup relative to S3-Hive (faster than the baseline)
b)
More disk IO relative to S3-Hive (slower than the baseline)
c)
More data stored in HDFS relative to S3-Hive
d)
Higher query complexity measured by number of joins
e)
More workers failing during execution
68.
In the same chart, which query number shows the most pronounced speedup for HDFS-Presto compared to the other queries shown?
a)
Query 5
b)
Query 1
c)
Query 2
d)
Query 4
e)
Query 6
69.
The Netflix benchmark slide lists three workload types. Which mapping matches the slide annotations that include "1MR" and "2MR"?
a)
GroupBy: 1MR; Join + GroupBy: 2MR; Needle in Haystack: 1MR
b)
GroupBy: 2MR; Join + GroupBy: 1MR; Needle in Haystack: 2MR
c)
GroupBy: 1MR; Join + GroupBy: 1MR; Needle in Haystack: 2MR
d)
GroupBy: 2MR; Join + GroupBy: 2MR; Needle in Haystack: 1MR
e)
All three are labeled 2MR
70.
The Netflix benchmark setup describes input and cluster resources. Which configuration matches the slide?
a)
Parquet input files on S3 (about 140GB to 210GB) on a 40-node m2.4xlarge cluster
b)
ORC input files on HDFS (about 10GB to 20GB) on a 4-node m1.large cluster
c)
RCFile input files on local disks (about 1TB) on a single server
d)
Sequence files on S3 (about 5GB) on a 10-node cluster
e)
Parquet input files on Cassandra on a 1000-node cluster
71.
In the "Presto read performance" chart, which format family is consistently the slowest across the different column-selection cases?
a)
Sequence files (on S3 and HDFS)
b)
ORC files (on S3 and HDFS)
c)
Parquet files (on S3 and HDFS)
d)
ORC and Parquet are always slower than Sequence
e)
All formats are shown with identical execution times
72.
The same chart compares selecting few columns versus many columns (including SELECT *). Which pattern is most consistent with the plotted results?
a)
Execution time generally increases as more columns are selected, with ORC/Parquet remaining well below Sequence
b)
Execution time generally decreases as more columns are selected, with Sequence improving the most
c)
Execution time is constant across column counts for all formats and storage systems
d)
ORC becomes slower than Sequence specifically when selecting a single column
e)
Parquet is only fast on HDFS but becomes the slowest on S3 for all selections
73.
In the Cloudera benchmark (single user response time), which system shows the lowest response time across Interactive, Reporting, and Deep Analytics in the plotted bars?
a)
Impala/Parquet
b)
Presto/RCFile
c)
Hive-on-Tez/ORC
d)
Shark/ORC (HDFS)
e)
Shark/ORC (RDD)
74.
In the Cloudera benchmark comparing single user versus 10 users, which system exhibits the smallest increase in response time when moving from single-user to 10-user load?
a)
Impala/Parquet
b)
Presto/RCFile
c)
Hive-on-Tez/ORC
d)
Shark/ORC (HDFS)
e)
Shark/ORC (RDD)
75.
In the ORC raw decoding speedup chart comparing an old Hive-based reader versus a new Presto ORC reader, which case shows the largest speedup for the "double" category?
a)
Uncompressed
b)
Snappy
c)
Zlib
d)
All three are identical for double
e)
None of the three show any speedup
Reset
