SSD Nodes Learn Hosting plans →
How to do am Matt ConnorBy Matt Connor

Sentry Alternative Wey Fit Run With 512 MB RAM

Sentry feature-complete setup ask for 16 GB RAM plus 16 GB swap, while GlitchTip documents 512 MB. Compare disk growth, containers and upgrade pain.

Wet self-hosted error tracking dey cost before e store one event

Self-hosted error tracking get one number wey dey decide the whole choice: na the RAM floor. Sentry own self-hosted documentation ask for 4 CPU cores, 16 GB of RAM plus 16 GB of swap, and 20 GB of free disk for the feature-complete stack, before your application send even one event. E get lower installer thresholds for errors-only profile: 2 CPU cores and 7000 MB of RAM available to Docker. GlitchTip document 512 MB. All the options here fit accept events from the same Sentry SDKs, so this no be decision about how you instrument your code. Na decision about how big server you dey ready to pay for and keep running.

Figures wey each project publish about its resources, side by side

These na the numbers wey each project publish about itself, as of October 2026. Dem no be the same type of measurement, so read the note for each row before you compare dem.

ChartPublished RAM figures per error tracking stack, October 2026
Di data wey dey behind dis chart
[
  {
    "tool": "Sentry feature-complete",
    "published_ram_gb": 16,
    "notes": "documented minimum for the feature-complete profile, plus 16 GB of swap"
  },
  {
    "tool": "Bugsink",
    "published_ram_gb": 4,
    "notes": "the vendor's own published benchmark box, 2 vCPU"
  },
  {
    "tool": "GlitchTip",
    "published_ram_gb": 0.5,
    "notes": "documented recommendation, 256 MB stated as the minimum"
  }
]

Sentry 16 GB na documented minimum for the feature-complete profile, and the same page recommend 32 GB. GlitchTip 0.5 GB na recommendation, and the project talk say 256 MB na working minimum, or 128 MB plus swap if you configure am carefully. Bugsink 4 GB no be any of those: na the box wey the vendor use for its own throughput benchmark. Published figure na starting point; e no promise how your event volume go behave.

Sentry self-hosted: most of the product, and the whole bill

The official stack na getsentry/self-hosted, a Docker Compose project wey dey run the same components Sentry dey run for production. The documentation list wetin no dey inside, including Seer and other closed source AI and ML features, billing, spike protection, and some platform support. You go get most Sentry features, plus the moving parts wey make those features work.

Install from a tagged release instead of master:

VERSION=$(curl -Ls -o /dev/null -w %{url_effective} https://github.com/getsentry/self-hosted/releases/latest)
VERSION=${VERSION##*/}
git clone https://github.com/getsentry/self-hosted.git
cd self-hosted
git checkout ${VERSION}
./install.sh

The feature-complete profile na the default. To install the smaller errors-only stack, set COMPOSE_PROFILES=errors-only inside .env before you run ./install.sh.

Then start am:

docker compose up --wait

By default, Sentry publish port 9000 for all host interfaces; set SENTRY_BIND=127.0.0.1:9000 inside .env to bind am to localhost. Docker Engine 19.03.6 or later and Docker Compose 2.32.2 or later dey required, and the installer dey check both: older Compose go stop ./install.sh with one line wey start with FAIL: Expected minimum and name the version wey e find, then the script go exit with error. For x86-64 hosts, official ClickHouse 26.6 and later images still require the complete x86-64-v3 CPU feature set, including AVX2.

Check wetin you actually start:

docker compose ps
free -h

docker compose ps list every service for the stack, and the list long: Postgres, ClickHouse, Kafka, Valkey (under the Compose service name redis), Relay, Snuba, Symbolicator, plus several worker and cron processes. Count dem once, because that number na your maintenance load. Every entry na one process wey fit crash, fill disk, or fail migration.

If service remain for Restarting state, check memory before anything else:

dmesg -T | grep -i 'out of memory'

A line like Out of memory: Killed process 3412 (java) mean say the kernel OOM killer (out of memory killer) kill one container because the machine run out of RAM. So that service no go become healthy, and the stack no go finish starting. The installer still dey check available RAM and reject machine wey dey below the minimum wey e enforce. The documentation also warn about disk speed: iowait above 10% probably mean say the machine no fit keep up with the ingest pipeline. Read am from the wa column inside top, or from iostat -x 5 if sysstat dey installed.

Upgrades na the part wey people dey underestimate

Sentry self-hosted dey release monthly under CalVer, a calendar-based version scheme, with one primary release on the 15th of every month. You no fit jump from old version straight to the latest one. The project define hard stop versions, and you must check out each one in order so you fit apply its database migrations. As of October 2026, the published hard stops na 9.1.2, 21.5.0, 21.6.3, 23.6.2, 23.11.0, 24.8.0, 25.5.1, 26.5.0 and 26.7.0. The documentation also list releases wey you suppose skip because of migration problems, including 23.7.0, 25.9.0, 25.12.0 and the 26.3.0 to 26.4.0 range.

An upgrade na checkout plus another run of the installer:

git fetch
git checkout 26.7.0
./install.sh
docker compose up --wait

Take snapshot of the server before you start, because migration across large ClickHouse dataset fit run for hours, and failure halfway fit leave the database between two schemas. The plain reason why most self-hosted Sentry upgrades fail na this: the machine remain on one version for one year, so the jump cross several hard stops at once, and one of the skipped migrations na the one wey matter.

One more thing you need know before you commit. Sentry self-hosted dey under the Functional Source License (FSL), wey Sentry itself introduce. E na fair source, not OSI-approved open source: you fit run am for yourself, but you no fit sell am as competing service. Each release go convert to Apache 2.0 two years after e ship.

GlitchTip: na 512 MB answer

GlitchTip get MIT license and e dey receive events from Sentry open source SDKs. So, instrumented application fit move over by changing one value: the DSN (data source name, na the URL wey your SDK dey post events go). GlitchTip 7.0 need PostgreSQL 15 or later. Valkey or Redis 7 or later no compulsory, but e dey make larger instances faster.

Installation na Docker plus one compose file:

sudo apt install docker.io
curl -O https://glitchtip.com/assets/compose.sample.yml
mv compose.sample.yml compose.yml

Edit the environment section before you start anything. Set the secret and domain, and configure mail if you want email alerts and password resets:

SECRET_KEY: <output of openssl rand -base64 50>
GLITCHTIP_DOMAIN: https://errors.example.com
DEFAULT_FROM_EMAIL: errors@example.com
EMAIL_URL: smtp://user:password@smtp.example.com:587

Set DATABASE_URL to the PostgreSQL connection string for your installation. GLITCHTIP_DOMAIN must include the scheme. If https:// no dey for front, the links inside alert emails go build wrongly and land for URL wey no dey answer.

Start am and monitor the first boot:

docker compose up -d
docker compose logs -f web

Keep image tags pinned. Compose file wey talk say latest go fetch new database engine for the next docker compose pull and switch to am the next time docker compose up -d recreates the container. Postgres major version jump under existing data directory fit make working error tracker stop starting.

To reach 256 MB to 512 MB range, disable Valkey and the optional log and uptime features. If you run without Valkey, GlitchTip go use its database for cache and queue work instead. E slower, but e still correct. All in one mode dey run the worker inside the web process, so you go maintain one application container instead of two.

Put proxy for front of am. GlitchTip documentation ask for proxy or load balancer wey buffers requests and handles chunked Transfer-Encoding, and e use nginx as the worked example. Without buffering, slow client go hold application worker open for the whole upload. Because of this, few slow senders fit occupy every worker wey you get, and healthy clients go begin timeout.

Read the release notes before you upgrade across major versions. For GlitchTip 7.0, upgrade to 6.x first if you dey use 5.x. Upgrade PostgreSQL 14 while you still dey use GlitchTip 6.x. Check renamed environment variables, removed startup settings, the new TRUSTED_PROXIES default, and changed permissions before you change application image tag to 7. Then pull and restart:

docker compose pull
docker compose stop
docker compose up -d

Database migrations dey run automatically when e start. Still take a dump first, because automatic migration still be migration.

Bugsink: container wan, and licence wey you must read

Bugsink na the lightest among the three. E dey speak Sentry SDK protocol, and e dey run without message queue or any external service apart from database. SQLite na the default, while MySQL and PostgreSQL dey supported when the installation don outgrow am.

Use this temporary instance to see the interface before you commit:

docker pull bugsink/bugsink:latest
docker run \
  -e SECRET_KEY=PUT_AN_ACTUAL_RANDOM_SECRET_HERE_OF_AT_LEAST_50_CHARS \
  -e CREATE_SUPERUSER=admin@example.org:admin \
  -e PORT=8000 \
  -p 8000:8000 \
  bugsink/bugsink

Open http://localhost:8000/ and sign in with the address and password wey you pass for CREATE_SUPERUSER. That container go lose all data when dem remove am. For real instance, use the project's compose sample. E pair bugsink/bugsink:2 with postgres:17-alpine and set DATABASE_URL, BASE_URL and BEHIND_HTTPS_PROXY. Generate the secret correctly:

openssl rand -base64 50

BASE_URL must match the URL wey your users and SDKs really dey use, including the scheme. If you leave am at http://localhost:8000 for a server wey you reach through https://errors.example.com, every link inside notification email go point to host wey no resolve for the person wey dey read am. Set BEHIND_HTTPS_PROXY to true when nginx or Caddy dey terminate TLS (transport layer security) in front of am, and forward X-Forwarded-Proto: https. This one make Bugsink recognize secure requests and make its CSRF checks use the correct scheme.

The vendor publish its own throughput figures: 18 events per second at 50 KB each. That one equal 1.5 million events per day on a 2 vCPU and 4 GB VPS. Treat this as the tool's general capacity, not as guarantee for your workload. E show say the limit dey far above wetin one small application fit produce.

Now na the licence. You need read this part before you add am to your stack. Bugsink dey released under PolyForm Shield License 1.0.0. That one mean source available, but e no be open source: you fit run am and modify am, but you no fit use am build something wey go compete with Bugsink. For internal error tracker, this restriction no go matter. If your company dey sell developer tooling, make person read the licence text first.

Error tracking and LLM observability na still two different tools

Search for one tool wey fit do error tracking and large language model (LLM) observability together, and you go see products wey claim say dem fit do both. The data structure no be the same, na why the merger never really work. Error tracker dey receive exception with stack trace, calculate fingerprint from am, then group thousands of occurrences into one issue with counter. LLM tracing tool dey receive span wey hold prompt, response, token count, and latency. E must keep every one of dem, because two calls with identical inputs still be separate events wey worth reading.

So run both. Send exceptions go error tracker, and send model calls go somewhere wey dem build for that purpose: self-hosted Langfuse for agent tracing cover that side, while self-hosted AI observability approach the same work from another angle. Your application already dey produce both kinds of failure. Model call wey return confident nonsense no throw any exception, so error tracker no go ever show am to you.

Disk growth na the failure wey go find you later

Every error tracker na write-heavy database with input wey no get limit. Your application decide how much data e go write, and one new bug for code path wey dey run often fit produce one million events overnight.

GlitchTip publishes one figure wey you need plan against: instance wey dey handle one million events per month fit need 30 GB disk space. Your retention window decide how much of that ingested data you go keep.

Bugsink dey handle am from the other side. Instead of fixed quota, e applies retention algorithm based on event count and event age, and e exposes the limits directly: MAX_RETENTION_EVENT_COUNT budgets the total of configured project retention counts, while MAX_RETENTION_PER_PROJECT_EVENT_COUNT caps each project's setting. These na configuration guardrails. MAX_EVENT_AGE_DAYS supplies absolute age limit when you run bugsink-manage delete_old_events or bugsink-manage vacuum --old-events. Set the event budget based on the actual sizes of events wey you store, because event count no be byte limit.

Monitor the real numbers for the server:

df -h /
docker system df -v
du -sh /var/lib/docker/volumes/*

docker system df -v prints the size for each volume, so you fit see which service dey grow. If volume dey gain several gigabytes every week without any change for traffic, e usually mean say retention window still dey fill up, or e dey hold more data than the partition fit contain at your event rate. If you leave retention as e be, that no mean say nothing dey expire: Sentry's shipped .env sets SENTRY_EVENT_RETENTION_DAYS=90, while GlitchTip's GLITCHTIP_RETENTION_DAYS defaults to 90. Reduce the window so e fit the disk instead of waiting for the partition to fill up.

Memory na the same problem, but e dey show for another way. Stack wey no get limits go take everything wey kernel offer, and when machine run out of memory, OOM killer go kill the process with the highest badness score. Kernel dey calculate this mainly from how much memory each process dey use, then e adjusts am with its oom_score_adj. That process fit be your web server instead of the tracker wey cause the problem. Give every service a maximum limit: memory limits for Docker Compose shows the syntax and wetin container dey do when e reach its limit. If container die because e reach its own limit, na contained failure be that. But when host memory finish, neighbouring services fit dey at risk too.

E stream fit which VPS

  • 1 GB, or 2 GB with room to spare: GlitchTip for all in one mode with Valkey off, or Bugsink for SQLite. Both dey comfortable here for small number of applications.
  • 4 GB: Bugsink with PostgreSQL, or GlitchTip with Valkey on and separate worker service. Na for this size you fit stop fine-tuning and just run am.
  • 8 GB: the official Sentry errors-only profile go pass installer's RAM check when Docker see at least 7000 MB; e also need 2 CPU cores. Set COMPOSE_PROFILES=errors-only for .env before installation.
  • 16 GB minimum, 32 GB recommended: the official Sentry feature-complete self-hosted stack, and only when you need Sentry feature wey the lighter projects no implement. Check the specific feature against each project's documentation first, because the compatible projects cover the common ones.

Anything wey you run, the error tracker no fit report say e don die by itself. Put check on am from another machine: Uptime Kuma wey dey watch from another box go tell you say the tracker dey down. Na exactly for that time your application go start dey throw errors wey nobody dey record.

When hosted plan na the cheaper answer

Self-hosting error tracker make sense when data residency rules require am, or when your event volume high enough say per-event pricing dey cost too much. Outside those cases, calculate the cost honestly. Sentry documented feature-complete minimum na server with 16 GB RAM, 4 cores, and fast disk. VPS wey get this size no be cheap VPS. Its errors-only profile get lower installer thresholds: 2 cores and 7000 MB Docker-visible RAM. Then add the operational work: crossing every hard stop in order, and taking snapshot before each migration, a few times every year.

GlitchTip and Bugsink change this calculation completely, because box with 512 MB to 4 GB RAM no expensive. Routine Docker updates na image pull followed by restart, after you check release notes and database requirements before major upgrades. Na why most people wey ask this question end up using one of the compatible projects instead of the official stack. Wetin dem want na error tracking, no be distributed data pipeline wey dem go dey monitor and maintain.

If you still dey work out wetin suppose dey on the server at all, the wider list of wetin worth self-hosting puts error tracking beside the other services wey dey compete for the same RAM.

FAQ

I fit self-host Sentry for 2 GB VPS?

No. Even Sentry errors-only profile need 2 CPU cores and at least 7000 MB RAM wey Docker get available; the feature-complete documentation talk say minimum na 4 CPU cores, 16 GB RAM plus 16 GB swap, and 20 GB free disk. The stack dey run Postgres, ClickHouse, Kafka, Valkey and several worker processes at the same time, and the installer dey reject 2 GB box during RAM check. If containers dey get killed, check dmesg -T | grep -i 'out of memory', wey go print one line naming the killed process. For 2 GB VPS, use GlitchTip, wey document 512 MB, or Bugsink, wey dey run as one container for SQLite.

I must change my application code to switch from Sentry go GlitchTip or Bugsink?

No. Both dey accept events from Sentry open source SDKs, so you fit keep the SDK wey you don already install and change one value: the DSN, wey be the URL wey SDK dey use post events. Move am go environment variable if e still hardcoded, point am to the new host, then raise test exception and monitor am as e arrive. If nothing show, check say project identifier for DSN match project wey dey exist for the new server, and say your firewall allow the application reach that host and port.

How much disk self-hosted error tracking need?

E depend on your event volume and retention window, no be on the tool itself. GlitchTip publish 30 GB for instance wey dey handle one million events per month. Bugsink let you budget configured project retention with MAX_RETENTION_EVENT_COUNT and apply MAX_EVENT_AGE_DAYS through its cleanup commands; disk sizing still depend on event sizes and database overhead. Check retention on the first day: Sentry and GlitchTip both ship with 90 day window, wey fit hold more data than your disk fit contain for your event rate. Tracker wey dey keep more data than disk fit hold go grow until df -h read 100%, and once e reach there, ingest go stop and you go lose the errors wey you most need see.

Why self-hosted Sentry upgrade dey keep fail?

Because the upgrade skip one hard stop. Sentry self-hosted define specific versions wey carry database migrations wey you must pass through, and as of October 2026, dem be 9.1.2, 21.5.0, 21.6.3, 23.6.2, 23.11.0, 24.8.0, 25.5.1, 26.5.0 and 26.7.0. If you go straight from old release to the newest one, you skip those migrations, so schema and code no agree and the upgrade go stop halfway. Check out each hard stop in order and run ./install.sh for every one, take server snapshot before you start, and read the documented list of releases to avoid, wey include 23.7.0, 25.9.0 and 25.12.0.

Dis article help you?

#error-tracking#sentry#glitchtip#observability#self-hosting