Question 690
An incident review requires ongoing per-process resource sampling, per-device disk latency measurements, and historical system activity. Which THREE tools match these separate needs? Choose THREE.
- Correlate application timing with block-device latency metrics
- pidstat -rud -p PID 1
- iostat -xz 1
- sysstat sar historical reports
- ss -tpn socket listing
- perf record and perf report
Explanation
Correct Answers
B:pidstat can sample process CPU, memory and I/O activity by PID over time. It provides more scope than one instantaneous process listing.
C:iostat -xz 1 samples extended block-device metrics, including throughput, utilization and request-latency indicators. It identifies the storage-device side of the incident alongside pidstat process samples and sar history.
D:With collection enabled, sar reports historical CPU, memory, disk and other system activity. It adds a retrospective timeline that a live pidstat process sample or current iostat device view cannot supply.
Incorrect Answers
A:Matching request latency with I/O latency, throughput and queue depth helps identify storage stalls; the evidence calls for capturing process metrics, block device latency and historic host trends, not merely this separate operation.
E:ss can associate TCP sockets with owning processes when permissions allow; the administrator instead needs capturing process metrics, block device latency and historic host trends; this option serves a different purpose.
F:Linux perf can sample hardware and software performance events to identify costly call paths; it answers a neighboring question rather than capturing process metrics, block device latency and historic host trends.
Question 691
The load spikes every evening and the operator needs historical system activity rather than current snapshots. Which facility can help?
- perf record and perf report
- ss -tpn socket listing
- sysstat sar historical reports
- Distributed tracing or request timing instrumentation
Explanation
Correct Answer
C:sar can retrieve time-series CPU, memory and I/O activity when collection is enabled. Historical data provides context for recurring spikes.
Incorrect Answers
A:Linux perf can sample hardware and software performance events to identify costly call paths; it answers a neighboring question rather than reviewing previously collected performance statistics.
B:ss can associate TCP sockets with owning processes when permissions allow; that changes a different part of the system; the requested outcome is reviewing previously collected performance statistics.
D:Request-level traces and latency spans reveal slow code paths and dependencies; it will not provide reviewing previously collected performance statistics under these conditions.
Question 692
A server's iostat shows sustained high await and low disk throughput during database reads. Which bottleneck is plausible?
- Network or NFS server I/O congestion
- High storage request latency
- Small synchronous write latency
- Disk I/O service capacity
Explanation
Correct Answer
B:Elevated await reflects time I/O requests spend queued and serviced. Low throughput does not rule out a heavily delayed device.
Incorrect Answers
A:Remote storage latency depends on both network transport and server disk service; the requirement is recognizing latency-bound storage requests, outside the effect of this choice.
C:Sequential megabytes per second do not characterize durability-sensitive small writes; the evidence calls for recognizing latency-bound storage requests, not merely this separate operation.
D:A busy storage device can accumulate latency as outstanding operations wait for service; the task involves recognizing latency-bound storage requests, whereas this addresses another subsystem.
Question 693
After adding physical RAM, the OS still reports the previous installed capacity. Which source should be reviewed?
- Firmware-reported memory and online memory blocks
- Resident set size (RSS)
- Container cgroup memory limit
- Memory allocation and OOM logs
Explanation
Correct Answer
A:The OS can only use memory that firmware and kernel successfully enumerate and online. Confirm DIMM health and memory block state.
Incorrect Answers
B:RSS measures physical memory currently resident for a process, though shared pages complicate accounting; the administrator instead needs finding why newly installed RAM is unavailable; this option serves a different purpose.
C:A memory cgroup can enforce an allocation ceiling independent of host capacity; it answers a neighboring question rather than finding why newly installed RAM is unavailable.
D:OOM messages show allocation context and selected processes; that changes a different part of the system; the requested outcome is finding why newly installed RAM is unavailable.
Question 694
An application writes a large log stream and the filesystem runs out of space unexpectedly. Which investigative approach isolates the consumer?
- Disk I/O service capacity
- Small synchronous write latency
- High storage request latency
- Compare df filesystem fullness with du directory sizes
- Idle CPU time with outstanding I/O
Explanation
Correct Answer
D:df shows allocated filesystem capacity while du attributes visible path consumption. Differences may indicate open deleted files or metadata.
Incorrect Answers
A:A busy storage device can accumulate latency as outstanding operations wait for service; it will not provide locating disk capacity consumers and discrepancies under these conditions.
B:Sequential megabytes per second do not characterize durability-sensitive small writes; using it would leave locating disk capacity consumers and discrepancies unresolved.
C:Elevated await reflects time I/O requests spend queued and serviced; the requirement is locating disk capacity consumers and discrepancies, outside the effect of this choice.
E:iowait reflects CPU idle intervals when at least one I/O request remains outstanding under Linux accounting; the evidence calls for locating disk capacity consumers and discrepancies, not merely this separate operation.
Question 695
NFS requests stall only during peak traffic, and retransmission counters rise. Which bottleneck class is plausible?
- Network or NFS server I/O congestion
- High storage request latency
- Idle CPU time with outstanding I/O
- IOPS and access pattern limit
Explanation
Correct Answer
A:Remote storage latency depends on both network transport and server disk service. Correlate mount statistics with network counters and backend load.
Incorrect Answers
B:Elevated await reflects time I/O requests spend queued and serviced; its scope differs from separating remote storage performance from local disk, so it cannot meet the stated requirement.
C:iowait reflects CPU idle intervals when at least one I/O request remains outstanding under Linux accounting; that is not equivalent to separating remote storage performance from local disk.
D:Random small operations stress per-request latency and IOPS rather than maximum sequential bandwidth; the administrator instead needs separating remote storage performance from local disk; this option serves a different purpose.
Question 696
df shows far more used blocks than du can account for, and a deleted log file remains open. What explains the discrepancy?
- Small synchronous write latency
- IOPS and access pattern limit
- Open deleted file retaining disk blocks
- High storage request latency
Explanation
Correct Answer
C:Unlinked files remain allocated while open file descriptors reference their inodes. Closing the process handle releases the blocks.
Incorrect Answers
A:Sequential megabytes per second do not characterize durability-sensitive small writes; that changes a different part of the system; the requested outcome is diagnosing disk blocks retained by deleted open files.
B:Random small operations stress per-request latency and IOPS rather than maximum sequential bandwidth; this does not achieve diagnosing disk blocks retained by deleted open files on the system described.
D:Elevated await reflects time I/O requests spend queued and serviced; using it would leave diagnosing disk blocks retained by deleted open files unresolved.
Question 697
A CPU profile reveals that one function dominates execution under load. Which optimization target is supported by this evidence?
- Correlate application timing with block-device latency metrics
- ss -tpn socket listing
- sysstat sar historical reports
- Optimize the measured hot function after correctness review
Explanation
Correct Answer
D:Profiling identifies where CPU time is spent, supporting targeted improvement. Unmeasured micro-optimizations may not improve end-to-end latency.
Incorrect Answers
A:Matching request latency with I/O latency, throughput and queue depth helps identify storage stalls; it does not deliver prioritizing optimization using measured hotspots, which is decisive in this scenario.
B:ss can associate TCP sockets with owning processes when permissions allow; the evidence calls for prioritizing optimization using measured hotspots, not merely this separate operation.
C:sar can retrieve time-series CPU, memory and I/O activity when collection is enabled; the task involves prioritizing optimization using measured hotspots, whereas this addresses another subsystem.
Question 698
A CPU-heavy service shows repeated throttling in its cgroup cpu.stat even while the host has idle cores. Its systemd unit has CPUQuota=20% set for a workload that needs more CPU. Which configuration should be reviewed?
- Increase the service CPUQuota= after checking capacity
- Increase the process nice value
- Enable additional swapping
- Decrease the open-file limit with ulimit -n
- Disable the system journal
Explanation
Correct Answer
A:systemd CPUQuota= controls the CPU time allotted to the service through its cgroup. If the configured quota is the bottleneck and capacity permits, adjusting it can reduce measured throttling.
Incorrect Answers
B:A higher nice value gives a process less favorable scheduler priority relative to competing tasks. It does not raise a hard cgroup CPU quota and may further reduce CPU access.
C:Swap affects the handling of memory pressure and cannot remove CPU time throttling enforced by the service cgroup. Host idle cores do not make swapping a solution.
D:The per-process file descriptor limit governs how many files or sockets a process can open. It cannot increase the CPU capacity allocated by CPUQuota=.
E:Journal logging changes neither the quota enforced by the CPU cgroup nor the observed throttling counters. Disabling logs would impair future incident diagnosis.
Question 699
Monitoring needs to distinguish anonymous memory from file-backed page cache when diagnosing a process. Which Linux source helps?
- Resident set size (RSS)
- /proc/PID/smaps_rollup
- Memory pressure causing disk-backed paging
- Reclaimable cache is not necessarily memory exhaustion
Explanation
Correct Answer
B:smaps_rollup provides summarized memory mappings and residency information for a process. It is more granular than a single overall system memory percentage.
Incorrect Answers
A:RSS measures physical memory currently resident for a process, though shared pages complicate accounting; using it would leave inspecting aggregated process memory mappings unresolved.
C:Swapping moves memory pages between RAM and backing storage, adding latency when working sets exceed available memory; it does not deliver inspecting aggregated process memory mappings, which is decisive in this scenario.
D:Linux uses spare RAM for caching and can reclaim much of it when applications demand space; the evidence calls for inspecting aggregated process memory mappings, not merely this separate operation.
Question 700
A service's resident memory rises steadily over several days and never returns after traffic falls. Which defect might be present?
- Firmware-reported memory and online memory blocks
- Memory leak
- Memory pressure causing disk-backed paging
- Page cache reclaim under memory pressure
- Resident set size (RSS)
Explanation
Correct Answer
B:A leak retains memory that is no longer needed, gradually increasing working-set pressure. Profiling and long-term process memory graphs can validate it.
Incorrect Answers
A:The OS can only use memory that firmware and kernel successfully enumerate and online; its scope differs from identifying unbounded retained memory, so it cannot meet the stated requirement.
C:Swapping moves memory pages between RAM and backing storage, adding latency when working sets exceed available memory; the administrator instead needs identifying unbounded retained memory; this option serves a different purpose.
D:Linux may reclaim page-cache pages to satisfy memory demands; it answers a neighboring question rather than identifying unbounded retained memory.
E:RSS measures physical memory currently resident for a process, though shared pages complicate accounting; that changes a different part of the system; the requested outcome is identifying unbounded retained memory.