HeartbeatHocusFocusFilesMacHyperTile
Guide

spindump, the daemon that writes down what a stuck Mac was doing

spindump is the macOS daemon that samples every process when an app hangs or blows a CPU or disk budget. Measured on a real Mac: 29 runs in 27 days, report headers quoted verbatim.

By Chethan Kumar S N · Updated

Every daemon in this series gets measured by the same ruler, and spindump finally gets to use one that fits. It is the first daemon in this series with its own manual page. Not a support article that never names it, not a decade old programming guide, but an actual spindump(8) entry in the System Manager’s Manual, one line of which does the whole job: “Profile entire system during a time interval.” After WindowServer with no man page and fseventsd with no documentation at all, the reporter is the one Apple documented properly. There is a reason for that, and it is the reason this page exists: spindump is the one daemon you are invited to run yourself.

The other daemons in this series run the Mac. spindump watches it fail.

What Apple says, in its own words

The manual page’s description section is three paragraphs, and the first explains every .diag file you have ever found on your disk:

spindump is used by various system components to create reports when an unresponsive application is force quit. Reports are stored at: /Library/Logs/DiagnosticReports/

For normal application force quits spindump will display a dialog to offer the choice to view more details and/or send a report to Apple.

That second sentence is the dialog you have seen, the one that appears after you force quit a beachballed app asking whether you want to view or send the report. The report it is offering you is spindump’s work product. The third paragraph describes the manual mode:

When run manually, spindump samples user and kernel callstacks for every process in the system.

Not the stuck process. Every process. The defaults are printed a few lines later: 10 seconds of sampling, one sample every 10 milliseconds, a hundred snapshots a second of the entire machine, kernel stacks included. When something hangs, macOS stops and takes attendance of everything running, then files it.

The identity card, measured

Same ruler as the rest of the series, all read from this Mac running macOS 27.0.1, 27 days of uptime:

The binary sits at /usr/sbin/spindump, 3,420,640 bytes, owned by root, built for arm64e and arm64e.x1, no Intel slice, signed with identifier com.apple.spindump and platform identifier 26, signed 9 August 2026. Unlike fseventsd, hidden two levels deep inside the framework it serves, spindump lives where a manual page says it lives, in the sbin directory, next to the other tools an administrator is meant to know about.

It runs from /System/Library/LaunchDaemons/com.apple.spindump.plist, as root, parented by launchd. The plist is short and interesting. It registers three Mach services: com.apple.spindump itself, plus com.apple.spinreporterd and com.apple.libquitd, two colleagues we will meet below. It sets EnablePressuredExit to true, the opposite of fseventsd’s setting, telling the system this daemon’s memory is fair game for reclaiming whenever it is idle. And it sets the environment pair the series keeps meeting, OSLogRateLimit 64 and MallocSpaceEfficient 1, keep the logging cheap and the footprint small.

The schedule is the best part. launchctl print reports runs = 29 over 27 days of uptime, which is almost exactly one run a day, and the immediate reason for the latest start was “ipc (mach)”. Not a timer. A message. Something woke it, which fits the job: you do not schedule a reporter, you dispatch one.

How a report gets born

Because the daemon runs about once a day, you can sit and wait for it. While this page was being written, a news app called NewsToday2 blew a disk budget, and the unified log caught the whole sequence inside 90 seconds.

At 12:52 the resident instance, process id 83869, 17 hours and 45 minutes old and idling at 4.8 MB of memory, woke up. First it went looking for names:

$ log show --last 6h --predicate 'process == "spindump"'
2026-11-16 12:53:39.644 WARNING: Inspecting live process [13860]
    to get symbol owner for <private>

A raw sample is just numbered addresses. To turn addresses into function names, spindump talks to coresymbolicationd, the system’s symbolication service, and the log shows the connection activating by name. Then, in a detail that ties this series together, it checks signatures: SecKeyVerifySignature, SecTrustEvaluateIfNecessary, SecTrustCopyAppleTrust Anchors, over and over, one for each binary it is about to name. The daemon that verifies what software is allowed to do, trustd, gets a workout here too, because a report is only trustworthy if the code it names is the code that ran.

Then the save, in CrashReporterSupport’s own words:

2026-11-16 12:53:44.009 (CrashReporterSupport) Saved diag report
    for NewsToday2 version 12.0 to
    NewsToday2_2026-11-16-125250_Chethans-MacBook-Pro.diag

And finally the daemon grades its own evidence, in its own logging category, which is how you learn the name of the mechanism underneath everything:

2026-11-16 12:53:44.094 [com.apple.spindump:logging] NewsToday2 [25247]:
    disk writes: some microstackshots with errors: 0 out-of-order
    microstackshots, 40 microstackshots missing load infos, 0 bytes invalid

The event was “disk writes”. The evidence was “microstackshots”. Hold that word.

Microstackshots: the part nobody tells you about

The manual page has a section most people never reach, and it explains how a report can describe an event that finished before spindump was ever woken:

Microstackshots are gathered by the kernel to provide extremely lightweight sampling of single threads at a time.

The kernel, not spindump, is always sampling, one thread at a time, at a cost so small the manual calls it extremely lightweight. When a process misbehaves, the kernel has already been keeping the receipts, and spindump’s job is to collect them, name them, and file them. That is why the header of every resource report on this Mac says “Data Source: Microstackshots”, and why the log line above counts errors in microstackshots rather than in samples spindump took itself. The reporter arrived after the incident and still wrote down what happened, because the camera was already rolling.

You can see this in any report’s header. Here is one from this Mac’s folder, a disk writes report for logd, the system log daemon, quoted verbatim:

Event:            disk writes
Action taken:     none
Writes:           2147.63 MB of file backed memory dirtied over 14185
                  seconds (151.40 KB per second average), exceeding limit
                  of 24.86 KB per second over 86400 seconds

Read that limit again, because it is a real policy stated in plain numbers: logd gets a budget of about 24.86 KB of writes per second, measured over a day, which works out to roughly 2 GB a day of log writing before the system takes notice. This Mac’s logd blew through it in four hours of heavy logging while the machine ran a load average above 12.

And here is the processor version, from a file whose name ends in .cpu_resource.diag, written when installd, the software installer daemon, was doing its job during an update:

Event:            cpu usage
Action taken:     none
CPU:              90 seconds cpu time over 146 seconds (62% cpu average),
                  exceeding limit of 50% cpu over 180 seconds

Fifty per cent of a core for three minutes is the tripwire, and installd tripped it installing software, which is what installd is for. Note the two words beside both events: “Action taken: none”. These reports are not punishments. Nothing was killed or throttled. Something exceeded a budget, the system wrote it down, and moved on.

The evidence drawer

/Library/Logs/DiagnosticReports, the folder the manual page names, is the drawer all this lands in. On this Mac, a developer machine with a load average of 14 on 10 cores while this page was written, the drawer held 26 spindump reports from the previous seven days, 6.7 MB of them, an average of 259 KB per report. The filenames read like a week of work: agent22, mds_stores, diskimages-helper, three separate rustc runs, bun, ditto, installd twice, ChatGPT, Claude, Claude Helper, Slack, Cisco Secure Client, CoreSimulatorBridge twice, and the biggest file of the lot, 1.8 MB, belonged to mds_stores, the Spotlight indexer, caught mid indexing storm on its own page’s territory.

Plain .diag files are the generic case, a hang or a disk writes event. The .cpu_resource.diag suffix marks the processor budget reports. All of them are plain text, all open in TextEdit, and all of them begin with the same header block: date, OS build, architecture, the command, its path, its code signing identifier, its parent, its process id, and then the Event and the numbers. You never need a tool to read what macOS recorded about your own machine.

The neighbourhood

The plist registered three Mach services, and the family is bigger than the daemon. spinreporterd takes the same name, the reporter that files reports when processes spin. tailspind, at /usr/libexec/tailspind, is the flight recorder variant: its launch configuration wakes it on notifications, a change of game mode, a configuration profile change, a tasking change, and it holds HostSpecialPort 13, one of the machine’s reserved ports. And Spotlight fields its own member: com.apple.metadata.mds.spindump, which despite the name is an mdworker running in MDSSpinDumpWorker mode, so the indexer can spin watch its own workers. When something hangs on this Mac, there is a whole agency that shows up, and spindump is the one who writes the article.

Activity Monitor knows all of this. Its toolbar has a Spindump button, and Apple describes it in exactly the manual page’s terms, creating a report for unresponsive apps that were terminated using force quit. The WindowServer page covers its sibling button, Sample Process, and the same warning applies: these are diagnostic tools for when something is wrong, and the wronger your machine is, the more of them you will meet.

Can you turn it off

No, and for once there is genuinely nothing to want. The reports are not overhead worth chasing, they are files, small, occasional, self pruning. The thing to fix is whatever fills the drawer: the app that hangs, the sync client that dirties two gigabytes, the build that pegs cores. The daemon itself costs almost nothing between incidents, and it exits on its own when idle, which this page caught happening live: the 17 hour old instance was gone six minutes after its last report, reclaimed under memory pressure exactly as EnablePressuredExit promised, with launchd showing state not running and runs still at 29. A reporter with nothing to report goes home.

Two commands if you want to watch the machinery yourself, both read only:

$ log stream --predicate 'process == "spindump"' --style compact

Then force quit any app that hangs and watch the save line arrive. And the drawer:

$ ls -lt /Library/Logs/DiagnosticReports | head

The short version

spindump is macOS’s incident reporter, the only daemon in this series with its own manual page, because it is the only one you are meant to run yourself. It samples every process on the machine, a hundred times a second for ten seconds, when an app is force quit after hanging, and it files reports about processes that exceed processor or disk budgets, budgets stated in the report headers in plain numbers, 50 per cent of a core for three minutes, 24.86 KB of writes per second over a day. The kernel does the always on watching through microstackshots so light the manual calls them extremely lightweight, and spindump does the naming, the signature checking, and the filing. On a heavily loaded Mac it wrote 26 reports in a week without once interfering. If you want to know what your Mac does when something breaks, the answer has been sitting in a folder the whole time, written by this.

Questions

What is spindump on Mac?

It is the daemon that writes the reports macOS keeps about misbehaving software. When an application stops responding and you force quit it, or when a process exceeds a processor or disk budget the system sets, spindump samples what every process on the machine was doing and saves the findings as a file in /Library/Logs/DiagnosticReports, with a name ending in .diag. Apple's own manual page, spindump(8), describes it as used by various system components to create reports when an unresponsive application is force quit. On the Mac measured here it ran as root, parented by launchd, and had started 29 times in 27 days of uptime, roughly once a day.

Why is spindump using so much CPU?

Almost always because you, or the system, just dealt with a stuck app. The manual page says that when run manually spindump samples user and kernel callstacks for every process in the system, by default for 10 seconds at 10 millisecond intervals, which is 100 samples a second across every process on the machine. That is real work while it lasts, and on a busy Mac spindump can itself appear near the top of Activity Monitor for those seconds. It is supposed to stop straight after: the launch configuration gives it a minimum runtime of 10 seconds, and on this Mac the resident instance idled at under one per cent of a core and then exited on its own six minutes after its last report. The one row that never leaves the Activity Monitor list it briefly tops is [sysmond](/guides/sysmond-mac), the daemon that gathers the statistics the app displays. spindump only visits. If spindump stays genuinely busy for minutes with no hang anywhere, the interesting question is what keeps asking it for reports.

What are the .diag files, and can I delete them?

They are spindump's finished reports, kept in /Library/Logs/DiagnosticReports alongside the crash files. The name carries the story: ChatGPT_2026-11-14-111114_Chethans-MacBook-Pro.diag, or installd_2026-11-10-110243 plus a .cpu_resource.diag suffix for processor budget reports. On this Mac, 26 of them had piled up in seven days, 6.7 MB in total, an average of 259 KB each. They are plain text, they open in any editor, and deleting them is safe: they are records of events that already happened, nothing on the system needs them to run, and macOS prunes the folder itself over time.

Is spindump a virus or something to worry about?

No. It is Apple's own code, and you can check that in five seconds: the real one lives at /usr/sbin/spindump, is signed with identifier com.apple.spindump and platform identifier 26, and runs as root under launchd, process id 1 as its parent. On the Mac measured here the binary was 3.4 MB, built for arm64e, dated August 2026. Malware can call itself anything, so if an activity named spindump is running from anywhere other than /usr/sbin, from a user account rather than root, that would be the moment to look closer. Nothing like that was present here.

Can I stop macOS from writing these reports?

There is no switch. The reports are not a service you can disable in System Settings, they are the paper trail of something else: an app that hung, a process that used more than 50 per cent of a processor for three minutes, a daemon that wrote two gigabytes to disk in a day. The honest fix is whatever the report describes, and you can read it yourself, because the header of every file states the event, the measurement and the limit in plain numbers. Deleting old reports is fine. If you are a developer, you can also run the sampler yourself with sudo spindump, which the manual page documents in full, and open the results it writes to /tmp.

Ready? Buy Ribbon for $15, or read all the guides.