Explain MVCC in PostgreSQL and how it affects concurrent transactions

Start by describing MVCC as PostgreSQL's approach to concurrency, where each transaction sees a consistent snapshot of data without blocking readers. Explain that rows have transaction metadata so readers can access older versions while writers create new versions. Give a short example: a long-running SELECT reads an earlier snapshot while an UPDATE creates a new row version, and VACUUM later removes dead tuples. Mention that this reduces lock contention for reads but increases space usage until cleanup runs. Tip: when answering, emphasize how MVCC trades write amplification for read concurrency and explain how you monitor bloat with pg_stat_all_tables. Avoid saying MVCC removes the need for any maintenance, and mention autovacuum tuning as a common follow-up.

What is WAL and how does it help with data durability?

Define WAL, the write ahead log, as PostgreSQL's sequential log of changes that must be written before data files are updated, which ensures durability. Explain how WAL supports crash recovery by replaying committed changes after an unclean shutdown. Give a practical example: enabling synchronous_commit affects when transactions wait for WAL to be flushed, trading latency for durability. Mention that WAL segments are used for replication and point-in-time recovery, and that wal_level and archive_mode settings control that behavior. Tip: clarify the difference between WAL flush and fsync, and avoid overstating performance; explain how WAL write throughput can become a bottleneck and how to monitor wal_writer activity.

How do indexes work in PostgreSQL and when would you use different index types?

Explain that indexes map key values to row locations and let the planner avoid full table scans, and mention common types like B-tree, Hash, GIN, and GiST. Describe typical uses: B-tree for equality and range queries, GIN for JSONB and full-text, and GiST for geometric types. Provide an example: create a B-tree on created_at for range queries and a GIN on a jsonb column for containment queries, then show how EXPLAIN can confirm index usage. Note that partial indexes and expression indexes are useful when only a subset of rows or transformed values are commonly queried. Tip: warn that indexes incur write overhead and can hurt bulk loads, and recommend checking index selectivity with pg_stats and running EXPLAIN ANALYZE before adding indexes.

Explain VACUUM, autovacuum, and how to prevent bloat

Describe VACUUM as the process that recovers space from dead row versions and updates visibility maps, and explain autovacuum as the automated background process that runs VACUUM and ANALYZE. Mention that VACUUM FULL rewrites tables and requires an exclusive lock, while plain VACUUM is non-blocking for reads. Give a technique: tune autovacuum_vacuum_scale_factor and autovacuum_vacuum_threshold for high-churn tables and schedule VACUUM FULL during maintenance windows only if necessary. Use monitoring metrics like pg_stat_user_tables.n_dead_tup and pg_total_relation_size to spot bloat before it affects performance. Tip: when answering, show you know to balance autovacuum aggressiveness with system load, and avoid suggesting VACUUM FULL as a first response to bloat without understanding root causes.

How does the PostgreSQL query planner work and how do you interpret EXPLAIN ANALYZE?

Explain that the planner builds possible query plans and estimates costs based on statistics, then the executor runs the chosen plan, and EXPLAIN ANALYZE shows actual runtimes and row counts. Emphasize the difference between estimated and actual rows as the key to spotting bad estimates. Give an approach: run EXPLAIN (FORMAT TEXT) to see the plan shape, then EXPLAIN ANALYZE to compare estimates and adjust with ANALYZE or stats adjustments, and check for sequential scans, nested loops, or misused indexes. Mention using pg_stat_statements to find hotspots and testing alternative plans with set enable_seqscan = off as an experiment, not a permanent fix. Tip: point out common pitfalls like stale statistics and parameterized queries affecting plan stability, and recommend keeping statistics up-to-date and using extended statistics for correlated columns.

Describe transaction isolation levels in PostgreSQL and common anomalies

List PostgreSQL's levels: Read Committed and Repeatable Read, and note that Serializable is implemented via predicate locking and SSI starting in modern releases. Explain practical effects: Read Committed sees committed changes between statements, Repeatable Read gives a statement-level snapshot, and Serializable can abort transactions to prevent anomalies. Give an example: under Read Committed, a second transaction can update a row between your two SELECTs, while Repeatable Read would keep the original snapshot, potentially causing serialization failures only under Serializable. Advise how to handle serialization failures by retrying transactions in your application. Tip: avoid claiming isolation eliminates all concurrency issues, and mention that correct handling often means detecting serialization errors and implementing idempotent retries to keep the system reliable.

How does partitioning work in PostgreSQL and when should you use it?

Explain declarative partitioning as creating a parent table with child partitions that hold subsets of data, and say partitioning helps with data management, faster deletes, and partition pruning. Mention range and list partitioning as common strategies and that partitioning can affect planning and vacuum behavior. Give a practical example: partition by month on a timestamp column to make dropping old partitions cheap and to reduce scanned rows for recent queries, and ensure your queries include the partition key so pruning happens. Note that too many partitions can add planning overhead and that maintenance tasks need to account for partitioned structures. Tip: when discussing, mention constraints like foreign key limitations on partitions in older versions and suggest testing query patterns to confirm pruning works as expected.

Explain streaming replication vs logical replication in PostgreSQL

Define streaming replication as physical WAL shipping that creates near-real-time standby replicas with identical data files, and logical replication as decoding changes into a logical stream that can be applied to different schemas or selective tables. Explain that streaming is typically used for high-availability read replicas, while logical fits migrations or selective replication scenarios. Give an example: use streaming replication for a hot standby ready for failover and use logical replication to replicate a subset of tables to a reporting database or between major version upgrades. Mention configuration points like wal_level, max_wal_senders, and publication/subscription setup for logical replication. Tip: clarify failover and replication lag monitoring, and avoid suggesting logical replication as a drop-in high-availability solution without addressing conflict resolution and setup complexity.

When would you choose JSONB over relational columns and how do you query it?

Explain that JSONB is useful when you need flexible schemas for semi-structured data, but that relational columns with clear types are better for frequent structured queries and constraints. Describe JSONB's advantages like indexing with GIN and fast containment queries using the @> operator. Give a query example: create a GIN index with CREATE INDEX ON table USING GIN (data jsonb_path_ops) and query with WHERE data @> '{"status": "active"}' to filter by a key. Recommend extracting frequently queried fields into separate columns to improve performance and allow proper indexing and constraints. Tip: warn against storing everything as JSONB to avoid losing schema guarantees, and suggest a hybrid approach where structured fields are columns and flexible attributes live in JSONB.

How do you optimize a slow query, step-by-step approach?

Outline a methodical approach: reproduce the issue, capture EXPLAIN ANALYZE, check indexes and statistics, and profile I/O and CPU usage. Explain that solving slow queries usually requires isolating whether the bottleneck is planning, execution, or I/O. Give a concrete example: start by running EXPLAIN ANALYZE to see where time is spent, then add or adjust indexes, rewrite joins or subqueries for better cardinality, and consider materialized views for repeated expensive aggregations. Mention practical tools such as pg_stat_statements, auto_explain, and pgbadger for query visibility. Tip: emphasize making incremental changes and measuring each step, and avoid premature broad changes like dropping indexes without confirming impact on other queries.

Tell me about a time you optimized a critical slow query

Situation: On a reporting system, nightly reports started taking four times longer after a data volume increase and caused downstream delays. The reports ran complex joins across several large tables and impacted business dashboards. Task: You were asked to reduce the runtime so reports completed within the maintenance window without disrupting OLTP workloads. The goal was to identify root causes and implement a fix that scaled with data growth. Action: You captured EXPLAIN ANALYZE for the slow report and discovered a nested loop join reading many rows due to missing statistics and an absent index on a filtering column. You added a selective index, updated statistics with ANALYZE, and rewrote part of the query into a CTE to materialize an intermediate result for reuse. Result: The report runtime dropped from four hours to under 45 minutes, meeting the maintenance window and reducing load on the primary DB during business hours. You documented the findings and added monitoring alerts for similar regressions.

Describe a time you handled a production database outage

Situation: A primary database node failed unexpectedly during peak usage and failover did not trigger automatically due to a misconfiguration in the monitoring layer. The incident caused service degradation for several customer-facing APIs. Task: As the on-call DBA, you needed to restore service quickly, fail over to a standby, and identify why automated failover did not execute. You also had to coordinate with engineering to communicate status and mitigate customer impact. Action: You validated the last consistent WAL segment, performed a controlled promotion of the most current standby, and updated application connection strings to point to the new primary while limiting write traffic. After service recovery, you audited the monitoring and failover configuration, fixed the misconfiguration, and scripted a runbook for the team. Result: Service returned within the agreed SLA window and most customers saw minimal disruption, and the postmortem prevented recurrence by adding automatic checks and a tested failover playbook. You reduced recovery time in subsequent drills and improved team readiness.

Give an example of learning a new PostgreSQL feature and applying it to a project

Situation: Your team needed faster analytics on semi-structured event data and you researched JSONB improvements introduced in a recent PostgreSQL release. The product required quick turnaround for building summary dashboards without major schema changes. Task: You had to evaluate whether JSONB functions and GIN index improvements would give the needed performance and plan a safe rollout. The objective was to prototype and then apply the feature without disrupting existing queries. Action: You built a prototype using JSONB with GIN indexes and benchmarked common queries against a production-sized sample, comparing response times and index sizes. After successful tests, you migrated less critical attributes to JSONB, added appropriate indexes, and updated ORM mappings with clear documentation for developers. Result: Query latency for analytics dropped by a measurable margin in benchmarks and deployed dashboards met performance targets, allowing the team to deliver features faster without major schema changes. You established a pattern for future use of JSONB with clear guidelines.

How do you find and remove duplicate rows in PostgreSQL?

Explain the common pattern: identify duplicates with a grouping query that counts rows per key, then remove extras while keeping one copy, often using a CTE with row_number(). Provide a safe approach that first selects candidates before deleting so you can review them. Give a code example: WITH dup AS (SELECT id, row_number() OVER (PARTITION BY key_col ORDER BY id) rn FROM table) DELETE FROM table WHERE id IN (SELECT id FROM dup WHERE rn > 1); This keeps the first row and deletes subsequent duplicates. Tip: run the SELECT part alone to verify targets before DELETE and consider adding a unique constraint after cleanup to prevent recurrence; avoid running deletes without a tested backup and transaction strategy.

Show how to perform an upsert in PostgreSQL

Describe INSERT ... ON CONFLICT as PostgreSQL's upsert mechanism, which inserts a row or updates it when a conflict on a unique constraint occurs. Explain how to specify the conflict target and the SET clause to control the update behavior. Give an example: INSERT INTO users (id, email, last_seen) VALUES (1, 'a@example.com', now()) ON CONFLICT (id) DO UPDATE SET last_seen = EXCLUDED.last_seen; This updates last_seen when the id already exists. You can also DO NOTHING to skip on conflict. Tip: be cautious about race conditions if your update logic depends on current DB values, and consider using RETURNING to get the affected row and confirm the operation outcome.

How does partition pruning work and give an example

Explain partition pruning as the planner's ability to exclude partitions that cannot contain matching rows based on query constraints, which reduces scanned data. State that pruning can occur at plan time or execution time depending on whether planner can see constant expressions or parameters. Give an example: partition a table by range on created_at and query WHERE created_at BETWEEN '2025-01-01' AND '2025-01-31'; the planner prunes to only the monthly partitions that overlap that range. Emphasize ensuring queries include the partition key to enable pruning. Tip: test pruning with EXPLAIN to confirm partitions are excluded, and avoid patterns that hide the partition key, such as wrapping it in functions that prevent constant comparison.

How do you index JSONB for fast queries and what are common pitfalls?

Explain that GIN indexes are the standard for JSONB containment and existence queries, while expression indexes can cover specific JSONB fields for equality checks. Mention the choice between jsonb_ops and jsonb_path_ops for different query shapes and the trade-offs in index size and performance. Give an example: CREATE INDEX ON events USING GIN (payload jsonb_path_ops); and query WHERE payload @> '{"type": "click"}' to benefit from the index. For frequent key lookups, create expression indexes like CREATE INDEX ON events ((payload->>'user_id')). Tip: be aware GIN indexes can be large and slower to update, and avoid indexing highly volatile JSONB fields; consider partial indexes to limit size and improve write performance.

Explain row-level security and show a simple policy example

Describe row-level security (RLS) as a way to enforce per-row access control policies in the database, evaluated for each query under the table's security context. Explain policies are defined per table and PostgreSQL checks them after normal permission checks. Give an example: ALTER TABLE documents ENABLE ROW LEVEL SECURITY; CREATE POLICY user_policy ON documents USING (owner_id = current_setting('app.user_id')::int); This ensures queries only return rows owned by the current application user when the setting is applied. Remember to set the app.user_id at session start. Tip: avoid assuming RLS replaces application checks; test thoroughly and document how application sessions set context so policies behave as expected, and remember superusers bypass RLS by default.

How do you perform a base backup and set up a standby for streaming replication?

Outline steps: configure wal_level and max_wal_senders on primary, create a base backup with pg_basebackup or file system snapshot, and set up recovery.conf or standby.signal with primary_conninfo on the standby. Emphasize ensuring consistent WAL shipping and proper archive settings if needed for PITR. Give a concise command example: on the standby, run pg_basebackup -h primary -D /var/lib/postgresql/data -P -X stream; then create standby.signal and set primary_conninfo in postgresql.conf to connect to the primary. Monitor replication with pg_stat_replication on the primary. Tip: test failover procedures and ensure firewall and authentication (pg_hba.conf) allow replication connections, and avoid skipping WAL archive and backup verification steps before relying on a standby for recovery.

postgresql Interview Questions: Complete Guide

postgresql interview questions often cover concurrency, performance, replication, and practical SQL skills, and interviews can include whiteboard problems, live coding, and system design discussions. You should expect a mix of conceptual questions and hands-on tasks that test how you debug and optimize real-world queries. Stay calm, show your thought process, and be ready to explain trade-offs.

Common Interview Questions

Behavioral Questions (STAR Method)

STAR Method: Structure your answers using Situation, Task, Action, and Result to tell compelling stories about your experience.

Technical Questions

Questions to Ask the Interviewer

Show your interest by asking thoughtful questions

•What does success look like in this role after the first 6 months, specifically around database reliability and performance?
•Can you describe the current database architecture, including replication, backup strategy, and common pain points?
•How does the team measure and prioritize database performance work versus feature work?
•What recent incidents or performance regressions has the team faced and how were they addressed?
•How is schema change management handled, especially for large tables or high-traffic systems?

Interview Preparation Tips

1
Practice explaining your reasoning out loud while you read EXPLAIN plans, recruiters want to hear your thought process as much as the final answer.
2
Bring a short story about a concrete performance fix, including metrics before and after, to show measurable impact during interviews.
3
When given a live SQL problem, reproduce the issue locally if possible, run EXPLAIN ANALYZE, and propose incremental changes rather than a single big rewrite.
4
Prepare to discuss trade-offs, such as read latency versus durability settings, and be ready to recommend safe rollback plans for proposed changes.

postgresql Interview Questions: Complete Guide

Michael Rodriguez

Common Interview Questions

Behavioral Questions (STAR Method)

Technical Questions

Questions to Ask the Interviewer

Interview Preparation Tips

Related guides

agile Interview Questions: Complete Guide

algorithms Interview Questions: Complete Guide

angular Interview Questions: Complete Guide

aws Interview Questions: Complete Guide

postgresql Interview Questions: Complete Guide

Michael Rodriguez

Common Interview Questions

Q1Explain MVCC in PostgreSQL and how it affects concurrent transactions

Q2What is WAL and how does it help with data durability?

Q3How do indexes work in PostgreSQL and when would you use different index types?

Q4Explain VACUUM, autovacuum, and how to prevent bloat

Q5How does the PostgreSQL query planner work and how do you interpret EXPLAIN ANALYZE?

Q6Describe transaction isolation levels in PostgreSQL and common anomalies

Q7How does partitioning work in PostgreSQL and when should you use it?

Q8Explain streaming replication vs logical replication in PostgreSQL

Q9When would you choose JSONB over relational columns and how do you query it?

Q10How do you optimize a slow query, step-by-step approach?

Behavioral Questions (STAR Method)

B1Tell me about a time you optimized a critical slow query

B2Describe a time you handled a production database outage

B3Give an example of learning a new PostgreSQL feature and applying it to a project

Technical Questions

T1How do you find and remove duplicate rows in PostgreSQL?

T2Show how to perform an upsert in PostgreSQL

T3How does partition pruning work and give an example

T4How do you index JSONB for fast queries and what are common pitfalls?

T5Explain row-level security and show a simple policy example

T6How do you perform a base backup and set up a standby for streaming replication?

Questions to Ask the Interviewer

Interview Preparation Tips

Related guides

agile Interview Questions: Complete Guide

algorithms Interview Questions: Complete Guide

angular Interview Questions: Complete Guide

aws Interview Questions: Complete Guide